fix: improve error handling and logging in data generation pipeline

CORRECTNESS FIXES:
- json_loader: Add RollbackTransaction() and call it on exception instead of
  CommitTransaction(). Prevents partial data corruption on parse/disk errors.
- wikipedia_service: Fix invalid MediaWiki API parameter explaintext=true ->
  explaintext=1. Now returns plain text instead of HTML markup in contexts.
- helpers: Fix ParseTwoLineResponse filter to only remove known thinking tags
  (<think>, <reasoning>, <reflect>) instead of any <...> pattern. Prevents
  silently removing legitimate output like <username>content</username>.

RELIABILITY & DESIGN IMPROVEMENTS:
- load/main: Make n_ctx (context window size) configurable via --n-ctx flag
  (default 2048, range 1-32768) to support larger models like Qwen3-14B.
- generate_brewery: Prevent retry prompt growth by extracting location context
  into constant and using compact retry format (error + schema + location only).
  Avoids token truncation on final retry attempts.
- database: Fix data representativeness by changing QueryCities from
  ORDER BY name (alphabetic bias) to ORDER BY RANDOM() for unbiased sampling.
  Convert all SQLITE_STATIC to SQLITE_TRANSIENT to prevent use-after-free risks.

POLISH:
- infer: Advance sampling seed between generation calls to improve diversity
  across brewery and user generation.
- data_downloader: Remove unnecessary commit hash truncation; use full hash.
- json_loader: Fix misleading log message from "RapidJSON" to "Boost.JSON".

pipeline/.clang-format

@@ -0,0 +1,5 @@
+---
+BasedOnStyle: Google
+ColumnLimit: 80
+IndentWidth: 3
+...

pipeline/.clang-tidy

@@ -0,0 +1,17 @@
+---
+Checks: >
+  -*,
+  bugprone-*,
+  clang-analyzer-*,
+  cppcoreguidelines-*,
+  google-*,
+  modernize-*,
+  performance-*,
+  readability-*,
+  -cppcoreguidelines-avoid-magic-numbers,
+  -cppcoreguidelines-owning-memory,
+  -readability-magic-numbers,
+  -google-readability-todo
+HeaderFilterRegex: "^(src|includes)/.*"
+FormatStyle: file
+...

pipeline/.gitignore

@@ -0,0 +1,5 @@
+dist
+build
+data
+models
+*.gguf

pipeline/CMakeLists.txt

@@ -0,0 +1,115 @@
+cmake_minimum_required(VERSION 3.24)
+project(biergarten-pipeline)
+# =============================================================================
+# 1. GPU Detection
+# =============================================================================
+# GGML_CUDA / GGML_METAL are set here so that the llama.cpp FetchContent below
+# inherits them as cache variables before its CMakeLists.txt is processed.
+if(APPLE)
+    message(STATUS "[biergarten] Apple Silicon detected — enabling Metal acceleration.")
+    set(GGML_METAL ON CACHE BOOL "Enable Metal for Apple Silicon" FORCE)
+elseif(UNIX AND NOT APPLE)
+    find_package(CUDAToolkit QUIET)
+    if(CUDAToolkit_FOUND)
+        message(STATUS "[biergarten] NVIDIA GPU detected — enabling CUDA acceleration.")
+        set(GGML_CUDA ON CACHE BOOL "Enable CUDA for NVIDIA GPUs" FORCE)
+        # 'native' resolves to the exact SM version of the present GPU at configure time
+        # (e.g. sm_89 for RTX 2000 Ada). Change to a concrete arch list for cross-compilation.
+        set(CMAKE_CUDA_ARCHITECTURES native)
+    else()
+        message(STATUS "[biergarten] No NVIDIA GPU found — falling back to CPU.")
+    endif()
+endif()
+# =============================================================================
+# 2. Project-wide Settings
+# =============================================================================
+set(CMAKE_CXX_STANDARD 23)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+# =============================================================================
+# 3. Dependencies
+# =============================================================================
+include(FetchContent)
+# --- libcurl ------------------------------------------------------------------
+# Prefer the system package; the build will fail at link time if absent and
+# no system curl is found, so emit a fatal error early rather than a silent gap.
+find_package(CURL QUIET)
+if(NOT CURL_FOUND)
+    message(FATAL_ERROR
+        "[biergarten] libcurl not found. Install it via your package manager "
+        "(e.g. 'sudo dnf install libcurl-devel') or set CURL_ROOT.")
+endif()
+# --- llama.cpp ----------------------------------------------------------------
+# Pinned to a specific commit for reproducible builds.
+# To update: pick a new commit SHA from https://github.com/ggml-org/llama.cpp
+FetchContent_Declare(
+    llama-cpp
+    GIT_REPOSITORY https://github.com/ggml-org/llama.cpp.git
+    GIT_TAG        b8611
+)
+FetchContent_MakeAvailable(llama-cpp)
+# --- Boost (JSON + program_options) ------------------------------------------
+FetchContent_Declare(
+    boost
+    URL https://github.com/boostorg/boost/releases/download/boost-1.85.0/boost-1.85.0-cmake.tar.gz
+)
+FetchContent_MakeAvailable(boost)
+# --- spdlog -------------------------------------------------------------------
+FetchContent_Declare(
+    spdlog
+    GIT_REPOSITORY https://github.com/gabime/spdlog.git
+    GIT_TAG        v1.15.3
+)
+FetchContent_MakeAvailable(spdlog)
+# =============================================================================
+# 4. Sources
+# =============================================================================
+set(SOURCES
+    src/main.cpp
+    src/biergarten_data_generator.cpp
+    src/data_generation/llama/destructor.cpp
+    src/data_generation/llama/generate_brewery.cpp
+    src/data_generation/llama/generate_user.cpp
+    src/data_generation/llama/helpers.cpp
+    src/data_generation/llama/infer.cpp
+    src/data_generation/llama/load.cpp
+    src/data_generation/llama/load_brewery_prompt.cpp
+    src/data_generation/llama/set_sampling_options.cpp
+    src/data_generation/mock/data.cpp
+    src/data_generation/mock/deterministic_hash.cpp
+    src/data_generation/mock/generate_brewery.cpp
+    src/data_generation/mock/generate_user.cpp
+    src/data_generation/mock/load.cpp
+    src/json_handling/json_loader.cpp
+    src/web_client/curl_web_client.cpp
+    src/wikipedia/wikipedia_service.cpp
+)
+# =============================================================================
+# 5. Target
+# =============================================================================
+add_executable(${PROJECT_NAME}
+    ${SOURCES}
+)
+target_include_directories(${PROJECT_NAME} PRIVATE
+    includes
+    ${llama-cpp_SOURCE_DIR}/include
+    ${llama-cpp_SOURCE_DIR}/common
+)
+target_link_libraries(${PROJECT_NAME} PRIVATE
+    llama
+    boost_json
+    boost_program_options
+    spdlog::spdlog
+    CURL::libcurl
+)
+
+# =============================================================================
+# 6. Runtime Assets
+# =============================================================================
+# Make locations.json available in the build directory for runtime relative path
+# lookups (e.g. when running from ./build).
+configure_file(
+    ${CMAKE_SOURCE_DIR}/locations.json
+    ${CMAKE_BINARY_DIR}/locations.json
+    COPYONLY
+)

pipeline/README.md

@@ -0,0 +1,406 @@
+# Biergarten Pipeline
+
+A high-performance C++23 data pipeline for fetching, parsing, and storing geographic data (countries, states, cities) with brewery metadata generation capabilities. The system supports both mock and LLM-based (llama.cpp) generation modes.
+
+## Overview
+
+The pipeline orchestrates **four key stages**:
+
+1. **Download** - Fetches `countries+states+cities.json` from a pinned GitHub commit with optional local filesystem caching
+2. **Parse** - Streams JSON using Boost.JSON's `basic_parser` to extract country/state/city records without loading the entire file into memory
+3. **Store** - Inserts records into a file-based SQLite database with all operations performed sequentially in a single thread
+4. **Generate** - Produces brewery metadata or user profiles (mock implementation; supports future LLM integration via llama.cpp)
+
+## System Architecture
+
+### Data Sources and Formats
+
+- **Hierarchical Structure**: Countries array → states per country → cities per state
+- **Data Fields**:
+  - `id` (integer)
+  - `name` (string)
+  - `iso2` / `iso3` (ISO country/state codes)
+  - `latitude` / `longitude` (geographic coordinates)
+- **Source**: [dr5hn/countries-states-cities-database](https://github.com/dr5hn/countries-states-cities-database) on GitHub
+- **Output**: Structured SQLite file-based database (`biergarten-pipeline.db`) + structured logging via spdlog
+
+### Concurrency Model
+
+The pipeline currently operates **single-threaded** with sequential stage execution:
+
+1. **Download Phase**: Main thread blocks while downloading the source JSON file (if not in cache)
+2. **Parse & Store Phase**: Main thread performs streaming JSON parse with immediate SQLite inserts
+
+**Thread Safety**: While single-threaded, the `SqliteDatabase` component is **mutex-protected** using `std::mutex` (`dbMutex`) for all database operations. This design enables safe future parallelization without code modifications.
+
+## Core Components
+
+| Component                     | Purpose                                                                                         | Thread Safety                                | Dependencies                                  |
+| ----------------------------- | ----------------------------------------------------------------------------------------------- | -------------------------------------------- | --------------------------------------------- |
+| **BiergartenDataGenerator**   | Orchestrates pipeline execution; manages lifecycle of downloader, parser, and generator         | Single-threaded coordinator                  | ApplicationOptions, WebClient, SqliteDatabase |
+| **DataDownloader**            | HTTP fetch with curl; optional filesystem cache; ETag support and retries                       | Blocking I/O; safe for startup               | IWebClient, filesystem                        |
+| **StreamingJsonParser**       | Extends `boost::json::basic_parser`; emits country/state/city via callbacks; tracks parse depth | Single-threaded parse; callbacks thread-safe | Boost.JSON                                    |
+| **JsonLoader**                | Wraps parser; dispatches callbacks for country/state/city; manages WorkQueue lifecycle          | Produces to WorkQueue; safe callbacks        | StreamingJsonParser, SqliteDatabase           |
+| **SqliteDatabase**            | Manages schema initialization; insert/query methods for geographic data                         | Mutex-guarded all operations                 | SQLite3                                       |
+| **IDataGenerator** (Abstract) | Interface for brewery/user metadata generation                                                  | Stateless virtual methods                    | N/A                                           |
+| **LlamaGenerator**            | LLM-based generation via llama.cpp; configurable sampling (temperature, top-p, seed)            | Manages llama_model* and llama_context*      | llama.cpp, BreweryResult, UserResult          |
+| **MockGenerator**             | Deterministic mock generation using seeded randomization                                        | Stateless; thread-safe                       | N/A                                           |
+| **CURLWebClient**             | HTTP client adapter; URL encoding; file downloads                                               | cURL library bindings                        | libcurl                                       |
+| **WikipediaService**          | (Planned) Wikipedia data lookups for enrichment                                                 | N/A                                          | IWebClient                                    |
+
+## Database Schema
+
+SQLite file-based database with **three core tables** and **indexes for fast lookups**:
+
+### Countries
+
+```sql
+CREATE TABLE countries (
+  id INTEGER PRIMARY KEY,
+  name TEXT NOT NULL,
+  iso2 TEXT,
+  iso3 TEXT
+);
+CREATE INDEX idx_countries_iso2 ON countries(iso2);
+```
+
+### States
+
+```sql
+CREATE TABLE states (
+  id INTEGER PRIMARY KEY,
+  country_id INTEGER NOT NULL,
+  name TEXT NOT NULL,
+  iso2 TEXT,
+  FOREIGN KEY (country_id) REFERENCES countries(id)
+);
+CREATE INDEX idx_states_country ON states(country_id);
+```
+
+### Cities
+
+```sql
+CREATE TABLE cities (
+  id INTEGER PRIMARY KEY,
+  state_id INTEGER NOT NULL,
+  country_id INTEGER NOT NULL,
+  name TEXT NOT NULL,
+  latitude REAL,
+  longitude REAL,
+  FOREIGN KEY (state_id) REFERENCES states(id),
+  FOREIGN KEY (country_id) REFERENCES countries(id)
+);
+CREATE INDEX idx_cities_state ON cities(state_id);
+CREATE INDEX idx_cities_country ON cities(country_id);
+```
+
+## Architecture Diagram
+
+```plantuml
+@startuml biergarten-pipeline
+!theme plain
+skinparam monochrome true
+skinparam classBackgroundColor #FFFFFF
+skinparam classBorderColor #000000
+
+package "Application Layer" {
+  class BiergartenDataGenerator {
+    - options: ApplicationOptions
+    - webClient: IWebClient
+    - database: SqliteDatabase
+    - generator: IDataGenerator
+    --
+    + Run() : int
+  }
+}
+
+package "Data Acquisition" {
+  class DataDownloader {
+    - webClient: IWebClient
+    --
+    + Download(url: string, filePath: string)
+    + DownloadWithCache(url: string, cachePath: string)
+  }
+
+  interface IWebClient {
+    + DownloadToFile(url: string, filePath: string)
+    + Get(url: string) : string
+    + UrlEncode(value: string) : string
+  }
+
+  class CURLWebClient {
+    - globalState: CurlGlobalState
+    --
+    + DownloadToFile(url: string, filePath: string)
+    + Get(url: string) : string
+    + UrlEncode(value: string) : string
+  }
+}
+
+package "JSON Processing" {
+  class StreamingJsonParser {
+    - depth: int
+    --
+    + on_object_begin()
+    + on_object_end()
+    + on_array_begin()
+    + on_array_end()
+    + on_key(str: string)
+    + on_string(str: string)
+    + on_number(value: int)
+  }
+
+  class JsonLoader {
+    --
+    + LoadWorldCities(jsonPath: string, db: SqliteDatabase)
+  }
+}
+
+package "Data Storage" {
+  class SqliteDatabase {
+    - db: sqlite3*
+    - dbMutex: std::mutex
+    --
+    + Initialize(dbPath: string)
+    + InsertCountry(id: int, name: string, iso2: string, iso3: string)
+    + InsertState(id: int, countryId: int, name: string, iso2: string)
+    + InsertCity(id: int, stateId: int, countryId: int, name: string, lat: double, lon: double)
+    + QueryCountries(limit: int) : vector<Country>
+    + QueryStates(limit: int) : vector<State>
+    + QueryCities() : vector<City>
+    + BeginTransaction()
+    + CommitTransaction()
+    # InitializeSchema()
+  }
+
+  struct Country {
+    id: int
+    name: string
+    iso2: string
+    iso3: string
+  }
+
+  struct State {
+    id: int
+    name: string
+    iso2: string
+    countryId: int
+  }
+
+  struct City {
+    id: int
+    name: string
+    countryId: int
+  }
+}
+
+package "Data Generation" {
+  interface IDataGenerator {
+    + load(modelPath: string)
+    + generateBrewery(cityName: string, countryName: string, regionContext: string) : BreweryResult
+    + generateUser(locale: string) : UserResult
+  }
+
+  class LlamaGenerator {
+    - model: llama_model*
+    - context: llama_context*
+    - sampling_temperature: float
+    - sampling_top_p: float
+    - sampling_seed: uint32_t
+    --
+    + load(modelPath: string)
+    + generateBrewery(...) : BreweryResult
+    + generateUser(locale: string) : UserResult
+    + setSamplingOptions(temperature: float, topP: float, seed: int)
+    # infer(prompt: string) : string
+  }
+
+  class MockGenerator {
+    --
+    + load(modelPath: string)
+    + generateBrewery(...) : BreweryResult
+    + generateUser(locale: string) : UserResult
+  }
+
+  struct BreweryResult {
+    name: string
+    description: string
+  }
+
+  struct UserResult {
+    username: string
+    bio: string
+  }
+}
+
+package "Enrichment (Planned)" {
+  class WikipediaService {
+    - webClient: IWebClient
+    --
+    + SearchCity(cityName: string, countryName: string) : string
+  }
+}
+
+' Relationships
+BiergartenDataGenerator --> DataDownloader
+BiergartenDataGenerator --> JsonLoader
+BiergartenDataGenerator --> SqliteDatabase
+BiergartenDataGenerator --> IDataGenerator
+
+DataDownloader --> IWebClient
+CURLWebClient ..|> IWebClient
+
+JsonLoader --> StreamingJsonParser
+JsonLoader --> SqliteDatabase
+
+LlamaGenerator ..|> IDataGenerator
+MockGenerator ..|> IDataGenerator
+
+SqliteDatabase --> Country
+SqliteDatabase --> State
+SqliteDatabase --> City
+
+LlamaGenerator --> BreweryResult
+LlamaGenerator --> UserResult
+MockGenerator --> BreweryResult
+MockGenerator --> UserResult
+
+WikipediaService --> IWebClient
+
+@enduml
+```
+
+## Configuration and Extensibility
+
+### Command-Line Arguments
+
+Boost.Program_options provides named CLI arguments. Running without arguments displays usage instructions.
+
+```bash
+./biergarten-pipeline [options]
+```
+
+**Requirement**: Exactly one of `--mocked` or `--model` must be specified.
+
+| Argument        | Short | Type   | Purpose                                                         |
+| --------------- | ----- | ------ | --------------------------------------------------------------- |
+| `--mocked`      | -     | flag   | Use mocked generator for brewery/user data                      |
+| `--model`       | `-m`  | string | Path to LLM model file (gguf); mutually exclusive with --mocked |
+| `--cache-dir`   | `-c`  | path   | Directory for cached JSON (default: `/tmp`)                     |
+| `--temperature` | -     | float  | LLM sampling temperature 0.0-1.0 (default: `0.8`)               |
+| `--top-p`       | -     | float  | Nucleus sampling parameter 0.0-1.0 (default: `0.92`)            |
+| `--seed`        | -     | int    | Random seed: -1 for random (default: `-1`)                      |
+| `--help`        | `-h`  | flag   | Show help message                                               |
+
+**Note**: The data source is always pinned to commit `c5eb7772` (stable 2026-03-28) and cannot be changed.
+
+**Note**: When `--mocked` is used, any sampling parameters (`--temperature`, `--top-p`, `--seed`) are ignored with a warning.
+
+### Usage Examples
+
+```bash
+# Mocked generator (deterministic, no LLM required)
+./biergarten-pipeline --mocked
+
+# With LLM model
+./biergarten-pipeline --model ./models/llama.gguf --cache-dir /var/cache
+
+# Mocked with extra parameters provided (will be ignored with warning)
+./biergarten-pipeline --mocked --temperature 0.5 --top-p 0.8 --seed 42
+
+# Show help
+./biergarten-pipeline --help
+```
+
+## Building and Running
+
+### Prerequisites
+
+- **C++23 compiler** (g++, clang, MSVC)
+- **CMake** 3.20+
+- **curl** (for HTTP downloads)
+- **sqlite3** (database backend)
+- **Boost** 1.75+ (requires Boost.JSON and Boost.Program_options)
+- **spdlog** v1.11.0 (fetched via CMake FetchContent)
+- **llama.cpp** (fetched via CMake FetchContent for LLM inference)
+
+### Build
+
+```bash
+mkdir -p build
+cd build
+cmake ..
+cmake --build . --target biergarten-pipeline -- -j
+```
+
+### Run
+
+```bash
+./build/biergarten-pipeline
+```
+
+**Output**:
+
+- Console logs with structured spdlog output
+- Cached JSON file: `/tmp/countries+states+cities.json`
+- SQLite database: `biergarten-pipeline.db` (in output directory)
+
+## Code Quality and Static Analysis
+
+### Formatting
+
+This project uses **clang-format** with the **Google C++ style guide**:
+
+```bash
+# Apply formatting to all source files
+cmake --build build --target format
+
+# Check formatting without modifications
+cmake --build build --target format-check
+```
+
+### Static Analysis
+
+This project uses **clang-tidy** with configurations for Google, modernize, performance, and bug-prone rules (`.clang-tidy`):
+
+Static analysis runs automatically during compilation if `clang-tidy` is available.
+
+## Code Implementation Summary
+
+### Key Achievements
+
+✅ **Full pipeline implementation** - Download → Parse → Store → Generate
+✅ **Streaming JSON parser** - Memory-efficient processing via Boost.JSON callbacks
+✅ **Thread-safe SQLite wrapper** - Mutex-protected database for future parallelization
+✅ **Flexible data generation** - Abstract IDataGenerator interface supporting both mock and LLM modes
+✅ **Comprehensive CLI** - Boost.Program_options with sensible defaults
+✅ **Production-grade logging** - spdlog integration for structured output
+✅ **Build quality** - CMake with clang-format/clang-tidy integration
+
+### Architecture Patterns
+
+- **Interface-based design**: `IWebClient`, `IDataGenerator` abstract base classes enable substitution and testing
+- **Dependency injection**: Components receive dependencies via constructors (BiergartenDataGenerator)
+- **RAII principle**: SQLite connections and resources managed via destructors
+- **Callback-driven parsing**: Boost.JSON parser emits events to processing callbacks
+- **Transaction-scoped inserts**: BeginTransaction/CommitTransaction for batch performance
+
+### External Dependencies
+
+| Dependency | Version | Purpose                            | Type    |
+| ---------- | ------- | ---------------------------------- | ------- |
+| Boost      | 1.75+   | JSON parsing, CLI argument parsing | Library |
+| SQLite3    | -       | Persistent data storage            | System  |
+| libcurl    | -       | HTTP downloads                     | System  |
+| spdlog     | v1.11.0 | Structured logging                 | Fetched |
+| llama.cpp  | b8611   | LLM inference engine               | Fetched |
+
+to validate formatting without modifying files.
+
+clang-tidy runs automatically on the biergarten-pipeline target when available. You can disable it at configure time:
+
+cmake -DENABLE_CLANG_TIDY=OFF ..
+
+You can also disable format helper targets:
+
+cmake -DENABLE_CLANG_FORMAT_TARGETS=OFF ..

pipeline/includes/biergarten_data_generator.h

@@ -0,0 +1,141 @@
+#ifndef BIERGARTEN_PIPELINE_BIERGARTEN_DATA_GENERATOR_H_
+#define BIERGARTEN_PIPELINE_BIERGARTEN_DATA_GENERATOR_H_
+
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "data_generation/data_generator.h"
+#include "models/location.h"
+#include "web_client/web_client.h"
+#include "wikipedia/wikipedia_service.h"
+
+/**
+ * @brief Program options for the Biergarten pipeline application.
+ */
+struct ApplicationOptions {
+   /// @brief Path to the LLM model file (gguf format); mutually exclusive with
+   /// use_mocked.
+   std::string model_path;
+
+   /// @brief Use mocked generator instead of LLM; mutually exclusive with
+   /// model_path.
+   bool use_mocked = false;
+
+   /// @brief Directory for cached JSON and database files.
+   std::string cache_dir;
+
+   /// @brief LLM sampling temperature (0.0 to 1.0, higher = more random).
+   float temperature = 0.8f;
+
+   /// @brief LLM nucleus sampling top-p parameter (0.0 to 1.0, higher = more
+   /// random).
+   float top_p = 0.92f;
+
+   /// @brief Context window size (tokens) for LLM inference. Higher values
+   /// support longer prompts but use more memory.
+   uint32_t n_ctx = 2048;
+
+   /// @brief Random seed for sampling (-1 for random, otherwise non-negative).
+   int seed = -1;
+
+   /// @brief Git commit hash for database consistency (always pinned to
+   /// c5eb7772).
+   std::string commit = "c5eb7772";
+};
+
+/**
+ * @brief Main data generator class for the Biergarten pipeline.
+ *
+ * This class encapsulates the core logic for generating brewery data.
+ * It handles location loading, city enrichment, and brewery generation.
+ */
+class BiergartenDataGenerator {
+  public:
+   /**
+    * @brief Construct a BiergartenDataGenerator with injected dependencies.
+    *
+    * @param options Application configuration options.
+    * @param web_client HTTP client for downloading data.
+    */
+   BiergartenDataGenerator(const ApplicationOptions& options,
+                           std::shared_ptr<WebClient> web_client);
+
+   /**
+    * @brief Run the data generation pipeline.
+    *
+    * Performs the following steps:
+    * 1. Load curated locations from JSON
+    * 2. Initialize the generator (LLM or Mock)
+    * 3. Generate brewery data for sampled cities
+    *
+    * @return 0 on success, 1 on failure.
+    */
+   int Run();
+
+  private:
+   /// @brief Immutable application options.
+   const ApplicationOptions options_;
+
+   /// @brief Shared HTTP client dependency.
+   std::shared_ptr<WebClient> webClient_;
+
+   /**
+    * @brief Enriched city data with Wikipedia context.
+    */
+   struct EnrichedCity {
+      Location location;
+      std::string region_context;
+   };
+
+   /**
+    * @brief Initialize the data generator based on options.
+    *
+    * Creates either a MockGenerator (if no model path) or LlamaGenerator.
+    *
+    * @return A unique_ptr to the initialized generator.
+    */
+   std::unique_ptr<DataGenerator> InitializeGenerator();
+
+   /**
+    * @brief Load locations from JSON and sample cities.
+    *
+    * @return Vector of sampled locations capped at 30 entries.
+    */
+   std::vector<Location> QueryCitiesWithCountries();
+
+   /**
+    * @brief Enrich cities with Wikipedia summaries.
+    *
+    * @param cities Vector of sampled locations.
+    * @return Vector of enriched city data with context.
+    */
+   std::vector<EnrichedCity> EnrichWithWikipedia(
+       const std::vector<Location>& cities);
+
+   /**
+    * @brief Generate breweries for enriched cities.
+    *
+    * @param generator The data generator instance.
+    * @param cities Vector of enriched city data.
+    */
+   void GenerateBreweries(DataGenerator& generator,
+                          const std::vector<EnrichedCity>& cities);
+
+   /**
+    * @brief Log the generated brewery results.
+    */
+   void LogResults() const;
+
+   /**
+    * @brief Helper struct to store generated brewery data.
+    */
+   struct GeneratedBrewery {
+      Location location;
+      BreweryResult brewery;
+   };
+
+   /// @brief Stores generated brewery data.
+   std::vector<GeneratedBrewery> generatedBreweries_;
+};
+#endif  // BIERGARTEN_PIPELINE_BIERGARTEN_DATA_GENERATOR_H_

pipeline/includes/data_generation/data_generator.h

@@ -0,0 +1,29 @@
+#ifndef BIERGARTEN_PIPELINE_DATA_GENERATION_DATA_GENERATOR_H_
+#define BIERGARTEN_PIPELINE_DATA_GENERATION_DATA_GENERATOR_H_
+
+#include <string>
+
+struct BreweryResult {
+   std::string name;
+   std::string description;
+};
+
+struct UserResult {
+   std::string username;
+   std::string bio;
+};
+
+class DataGenerator {
+  public:
+   virtual ~DataGenerator() = default;
+
+   virtual void Load(const std::string& model_path) = 0;
+
+   virtual BreweryResult GenerateBrewery(const std::string& city_name,
+                                         const std::string& country_name,
+                                         const std::string& region_context) = 0;
+
+   virtual UserResult GenerateUser(const std::string& locale) = 0;
+};
+
+#endif  // BIERGARTEN_PIPELINE_DATA_GENERATION_DATA_GENERATOR_H_

pipeline/includes/data_generation/llama_generator.h

@@ -0,0 +1,51 @@
+#ifndef BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_H_
+#define BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_H_
+
+#include <cstdint>
+#include <string>
+
+#include "data_generation/data_generator.h"
+
+struct llama_model;
+struct llama_context;
+
+class LlamaGenerator final : public DataGenerator {
+  public:
+   LlamaGenerator() = default;
+   ~LlamaGenerator() override;
+
+   void SetSamplingOptions(float temperature, float top_p, int seed = -1);
+
+   void SetContextSize(uint32_t n_ctx);
+
+   void Load(const std::string& model_path) override;
+   BreweryResult GenerateBrewery(const std::string& city_name,
+                                 const std::string& country_name,
+                                 const std::string& region_context) override;
+   UserResult GenerateUser(const std::string& locale) override;
+
+  private:
+   std::string Infer(const std::string& prompt, int max_tokens = 10000);
+   // Overload that allows passing a system message separately so chat-capable
+   // models receive a proper system role instead of having the system text
+   // concatenated into the user prompt (helps avoid revealing internal
+   // reasoning or instructions in model output).
+   std::string Infer(const std::string& system_prompt,
+                     const std::string& prompt, int max_tokens = 10000);
+
+   std::string InferFormatted(const std::string& formatted_prompt,
+                              int max_tokens = 10000);
+
+   std::string LoadBrewerySystemPrompt(const std::string& prompt_file_path);
+   std::string GetFallbackBreweryPrompt();
+
+   llama_model* model_ = nullptr;
+   llama_context* context_ = nullptr;
+   float sampling_temperature_ = 0.8f;
+   float sampling_top_p_ = 0.92f;
+   uint32_t sampling_seed_ = 0xFFFFFFFFu;
+   uint32_t n_ctx_ = 8192;
+   std::string brewery_system_prompt_;
+};
+
+#endif  // BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_H_

pipeline/includes/data_generation/llama_generator_helpers.h

@@ -0,0 +1,32 @@
+#ifndef BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_HELPERS_H_
+#define BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_HELPERS_H_
+
+#include <string>
+#include <utility>
+
+struct llama_model;
+struct llama_vocab;
+typedef int llama_token;
+
+// Helper functions for LlamaGenerator methods
+std::string PrepareRegionContextPublic(std::string_view region_context,
+                                       std::size_t max_chars = 700);
+
+std::pair<std::string, std::string> ParseTwoLineResponsePublic(
+    const std::string& raw, const std::string& error_message);
+
+std::string ToChatPromptPublic(const llama_model* model,
+                               const std::string& user_prompt);
+
+std::string ToChatPromptPublic(const llama_model* model,
+                               const std::string& system_prompt,
+                               const std::string& user_prompt);
+
+void AppendTokenPiecePublic(const llama_vocab* vocab, llama_token token,
+                            std::string& output);
+
+std::string ValidateBreweryJsonPublic(const std::string& raw,
+                                      std::string& name_out,
+                                      std::string& description_out);
+
+#endif  // BIERGARTEN_PIPELINE_DATA_GENERATION_LLAMA_GENERATOR_HELPERS_H_

pipeline/includes/data_generation/mock_generator.h

@@ -0,0 +1,28 @@
+#ifndef BIERGARTEN_PIPELINE_DATA_GENERATION_MOCK_GENERATOR_H_
+#define BIERGARTEN_PIPELINE_DATA_GENERATION_MOCK_GENERATOR_H_
+
+#include <string>
+#include <vector>
+
+#include "data_generation/data_generator.h"
+
+class MockGenerator final : public DataGenerator {
+  public:
+   void Load(const std::string& model_path) override;
+   BreweryResult GenerateBrewery(const std::string& city_name,
+                                 const std::string& country_name,
+                                 const std::string& region_context) override;
+   UserResult GenerateUser(const std::string& locale) override;
+
+  private:
+   static std::size_t DeterministicHash(const std::string& a,
+                                        const std::string& b);
+
+   static const std::vector<std::string> kBreweryAdjectives;
+   static const std::vector<std::string> kBreweryNouns;
+   static const std::vector<std::string> kBreweryDescriptions;
+   static const std::vector<std::string> kUsernames;
+   static const std::vector<std::string> kBios;
+};
+
+#endif  // BIERGARTEN_PIPELINE_DATA_GENERATION_MOCK_GENERATOR_H_

pipeline/includes/json_handling/json_loader.h

@@ -0,0 +1,16 @@
+#ifndef BIERGARTEN_PIPELINE_JSON_HANDLING_JSON_LOADER_H_
+#define BIERGARTEN_PIPELINE_JSON_HANDLING_JSON_LOADER_H_
+
+#include <string>
+#include <vector>
+
+#include "models/location.h"
+
+/// @brief Loads curated world locations from a JSON file into memory.
+class JsonLoader {
+  public:
+   /// @brief Parses a JSON array file and returns all location records.
+   static std::vector<Location> LoadLocations(const std::string& filepath);
+};
+
+#endif  // BIERGARTEN_PIPELINE_JSON_HANDLING_JSON_LOADER_H_

pipeline/includes/web_client/curl_web_client.h

@@ -0,0 +1,30 @@
+#ifndef BIERGARTEN_PIPELINE_WEB_CLIENT_CURL_WEB_CLIENT_H_
+#define BIERGARTEN_PIPELINE_WEB_CLIENT_CURL_WEB_CLIENT_H_
+
+#include <memory>
+
+#include "web_client/web_client.h"
+
+// RAII for curl_global_init/cleanup.
+// An instance of this class should be created in main() before any curl
+// operations and exist for the lifetime of the application.
+class CurlGlobalState {
+  public:
+   CurlGlobalState();
+   ~CurlGlobalState();
+   CurlGlobalState(const CurlGlobalState&) = delete;
+   CurlGlobalState& operator=(const CurlGlobalState&) = delete;
+};
+
+class CURLWebClient : public WebClient {
+  public:
+   CURLWebClient();
+   ~CURLWebClient() override;
+
+   void DownloadToFile(const std::string& url,
+                       const std::string& file_path) override;
+   std::string Get(const std::string& url) override;
+   std::string UrlEncode(const std::string& value) override;
+};
+
+#endif  // BIERGARTEN_PIPELINE_WEB_CLIENT_CURL_WEB_CLIENT_H_

pipeline/includes/web_client/web_client.h

@@ -0,0 +1,22 @@
+#ifndef BIERGARTEN_PIPELINE_WEB_CLIENT_WEB_CLIENT_H_
+#define BIERGARTEN_PIPELINE_WEB_CLIENT_WEB_CLIENT_H_
+
+#include <string>
+
+class WebClient {
+  public:
+   virtual ~WebClient() = default;
+
+   // Downloads content from a URL to a file. Throws on error.
+   virtual void DownloadToFile(const std::string& url,
+                               const std::string& file_path) = 0;
+
+   // Performs a GET request and returns the response body as a string. Throws
+   // on error.
+   virtual std::string Get(const std::string& url) = 0;
+
+   // URL-encodes a string.
+   virtual std::string UrlEncode(const std::string& value) = 0;
+};
+
+#endif  // BIERGARTEN_PIPELINE_WEB_CLIENT_WEB_CLIENT_H_

pipeline/includes/wikipedia/wikipedia_service.h

@@ -0,0 +1,27 @@
+#ifndef BIERGARTEN_PIPELINE_WIKIPEDIA_WIKIPEDIA_SERVICE_H_
+#define BIERGARTEN_PIPELINE_WIKIPEDIA_WIKIPEDIA_SERVICE_H_
+
+#include <memory>
+#include <string>
+#include <string_view>
+#include <unordered_map>
+
+#include "web_client/web_client.h"
+
+/// @brief Provides cached Wikipedia summary lookups for city and country pairs.
+class WikipediaService {
+  public:
+   /// @brief Creates a new Wikipedia service with the provided web client.
+   explicit WikipediaService(std::shared_ptr<WebClient> client);
+
+   /// @brief Returns the Wikipedia summary extract for city and country.
+   [[nodiscard]] std::string GetSummary(std::string_view city,
+                                        std::string_view country);
+
+  private:
+   std::string FetchExtract(std::string_view query);
+   std::shared_ptr<WebClient> client_;
+   std::unordered_map<std::string, std::string> cache_;
+};
+
+#endif  // BIERGARTEN_PIPELINE_WIKIPEDIA_WIKIPEDIA_SERVICE_H_

pipeline/locations.json

@@ -0,0 +1,902 @@
+[
+  {
+    "city": "Cape Town",
+    "state_province": "Western Cape",
+    "iso3166_2": "ZA-WC",
+    "country": "South Africa",
+    "iso3166_1": "ZA",
+    "latitude": -33.9249,
+    "longitude": 18.4241
+  },
+  {
+    "city": "Johannesburg",
+    "state_province": "Gauteng",
+    "iso3166_2": "ZA-GT",
+    "country": "South Africa",
+    "iso3166_1": "ZA",
+    "latitude": -26.2041,
+    "longitude": 28.0473
+  },
+  {
+    "city": "Durban",
+    "state_province": "KwaZulu-Natal",
+    "iso3166_2": "ZA-NL",
+    "country": "South Africa",
+    "iso3166_1": "ZA",
+    "latitude": -29.8587,
+    "longitude": 31.0218
+  },
+  {
+    "city": "Franschhoek",
+    "state_province": "Western Cape",
+    "iso3166_2": "ZA-WC",
+    "country": "South Africa",
+    "iso3166_1": "ZA",
+    "latitude": -33.9146,
+    "longitude": 19.1198
+  },
+  {
+    "city": "Nairobi",
+    "state_province": "Nairobi",
+    "iso3166_2": "KE-30",
+    "country": "Kenya",
+    "iso3166_1": "KE",
+    "latitude": -1.2921,
+    "longitude": 36.8219
+  },
+  {
+    "city": "Buenos Aires",
+    "state_province": "Buenos Aires City",
+    "iso3166_2": "AR-C",
+    "country": "Argentina",
+    "iso3166_1": "AR",
+    "latitude": -34.6037,
+    "longitude": -58.3816
+  },
+  {
+    "city": "Bariloche",
+    "state_province": "Río Negro",
+    "iso3166_2": "AR-R",
+    "country": "Argentina",
+    "iso3166_1": "AR",
+    "latitude": -41.1335,
+    "longitude": -71.3103
+  },
+  {
+    "city": "Bogotá",
+    "state_province": "Bogotá D.C.",
+    "iso3166_2": "CO-DC",
+    "country": "Colombia",
+    "iso3166_1": "CO",
+    "latitude": 4.711,
+    "longitude": -74.0721
+  },
+  {
+    "city": "Medellín",
+    "state_province": "Antioquia",
+    "iso3166_2": "CO-ANT",
+    "country": "Colombia",
+    "iso3166_1": "CO",
+    "latitude": 6.2442,
+    "longitude": -75.5812
+  },
+  {
+    "city": "São Paulo",
+    "state_province": "São Paulo",
+    "iso3166_2": "BR-SP",
+    "country": "Brazil",
+    "iso3166_1": "BR",
+    "latitude": -23.5505,
+    "longitude": -46.6333
+  },
+  {
+    "city": "Curitiba",
+    "state_province": "Paraná",
+    "iso3166_2": "BR-PR",
+    "country": "Brazil",
+    "iso3166_1": "BR",
+    "latitude": -25.4284,
+    "longitude": -49.2733
+  },
+  {
+    "city": "Rio de Janeiro",
+    "state_province": "Rio de Janeiro",
+    "iso3166_2": "BR-RJ",
+    "country": "Brazil",
+    "iso3166_1": "BR",
+    "latitude": -22.9068,
+    "longitude": -43.1729
+  },
+  {
+    "city": "Santiago",
+    "state_province": "Santiago Metropolitan",
+    "iso3166_2": "CL-RM",
+    "country": "Chile",
+    "iso3166_1": "CL",
+    "latitude": -33.4489,
+    "longitude": -70.6693
+  },
+  {
+    "city": "Valdivia",
+    "state_province": "Los Ríos",
+    "iso3166_2": "CL-LR",
+    "country": "Chile",
+    "iso3166_1": "CL",
+    "latitude": -39.8142,
+    "longitude": -73.2459
+  },
+  {
+    "city": "Lima",
+    "state_province": "Lima",
+    "iso3166_2": "PE-LMA",
+    "country": "Peru",
+    "iso3166_1": "PE",
+    "latitude": -12.0464,
+    "longitude": -77.0428
+  },
+  {
+    "city": "Tokyo",
+    "state_province": "Tokyo",
+    "iso3166_2": "JP-13",
+    "country": "Japan",
+    "iso3166_1": "JP",
+    "latitude": 35.6762,
+    "longitude": 139.6503
+  },
+  {
+    "city": "Osaka",
+    "state_province": "Osaka",
+    "iso3166_2": "JP-27",
+    "country": "Japan",
+    "iso3166_1": "JP",
+    "latitude": 34.6937,
+    "longitude": 135.5023
+  },
+  {
+    "city": "Kyoto",
+    "state_province": "Kyoto",
+    "iso3166_2": "JP-26",
+    "country": "Japan",
+    "iso3166_1": "JP",
+    "latitude": 35.0116,
+    "longitude": 135.7681
+  },
+  {
+    "city": "Sapporo",
+    "state_province": "Hokkaido",
+    "iso3166_2": "JP-01",
+    "country": "Japan",
+    "iso3166_1": "JP",
+    "latitude": 43.0618,
+    "longitude": 141.3545
+  },
+  {
+    "city": "Seoul",
+    "state_province": "Seoul",
+    "iso3166_2": "KR-11",
+    "country": "South Korea",
+    "iso3166_1": "KR",
+    "latitude": 37.5665,
+    "longitude": 126.978
+  },
+  {
+    "city": "Busan",
+    "state_province": "Busan",
+    "iso3166_2": "KR-26",
+    "country": "South Korea",
+    "iso3166_1": "KR",
+    "latitude": 35.1796,
+    "longitude": 129.0756
+  },
+  {
+    "city": "Ho Chi Minh City",
+    "state_province": "Ho Chi Minh",
+    "iso3166_2": "VN-SG",
+    "country": "Vietnam",
+    "iso3166_1": "VN",
+    "latitude": 10.8231,
+    "longitude": 106.6297
+  },
+  {
+    "city": "Hanoi",
+    "state_province": "Hanoi",
+    "iso3166_2": "VN-HN",
+    "country": "Vietnam",
+    "iso3166_1": "VN",
+    "latitude": 21.0285,
+    "longitude": 105.8542
+  },
+  {
+    "city": "Da Nang",
+    "state_province": "Da Nang",
+    "iso3166_2": "VN-DN",
+    "country": "Vietnam",
+    "iso3166_1": "VN",
+    "latitude": 16.0544,
+    "longitude": 108.2022
+  },
+  {
+    "city": "Bangkok",
+    "state_province": "Bangkok",
+    "iso3166_2": "TH-10",
+    "country": "Thailand",
+    "iso3166_1": "TH",
+    "latitude": 13.7563,
+    "longitude": 100.5018
+  },
+  {
+    "city": "Taipei",
+    "state_province": "Taipei",
+    "iso3166_2": "TW-TPE",
+    "country": "Taiwan",
+    "iso3166_1": "TW",
+    "latitude": 25.033,
+    "longitude": 121.5654
+  },
+  {
+    "city": "Beijing",
+    "state_province": "Beijing",
+    "iso3166_2": "CN-BJ",
+    "country": "China",
+    "iso3166_1": "CN",
+    "latitude": 39.9042,
+    "longitude": 116.4074
+  },
+  {
+    "city": "Shanghai",
+    "state_province": "Shanghai",
+    "iso3166_2": "CN-SH",
+    "country": "China",
+    "iso3166_1": "CN",
+    "latitude": 31.2304,
+    "longitude": 121.4737
+  },
+  {
+    "city": "Bengaluru",
+    "state_province": "Karnataka",
+    "iso3166_2": "IN-KA",
+    "country": "India",
+    "iso3166_1": "IN",
+    "latitude": 12.9716,
+    "longitude": 77.5946
+  },
+  {
+    "city": "Singapore",
+    "state_province": "Central Singapore",
+    "iso3166_2": "SG-01",
+    "country": "Singapore",
+    "iso3166_1": "SG",
+    "latitude": 1.3521,
+    "longitude": 103.8198
+  },
+  {
+    "city": "Melbourne",
+    "state_province": "Victoria",
+    "iso3166_2": "AU-VIC",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -37.8136,
+    "longitude": 144.9631
+  },
+  {
+    "city": "Sydney",
+    "state_province": "New South Wales",
+    "iso3166_2": "AU-NSW",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -33.8688,
+    "longitude": 151.2093
+  },
+  {
+    "city": "Brisbane",
+    "state_province": "Queensland",
+    "iso3166_2": "AU-QLD",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -27.4705,
+    "longitude": 153.026
+  },
+  {
+    "city": "Adelaide",
+    "state_province": "South Australia",
+    "iso3166_2": "AU-SA",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -34.9285,
+    "longitude": 138.6007
+  },
+  {
+    "city": "Perth",
+    "state_province": "Western Australia",
+    "iso3166_2": "AU-WA",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -31.9505,
+    "longitude": 115.8605
+  },
+  {
+    "city": "Hobart",
+    "state_province": "Tasmania",
+    "iso3166_2": "AU-TAS",
+    "country": "Australia",
+    "iso3166_1": "AU",
+    "latitude": -42.8821,
+    "longitude": 147.3272
+  },
+  {
+    "city": "Wellington",
+    "state_province": "Wellington",
+    "iso3166_2": "NZ-WGN",
+    "country": "New Zealand",
+    "iso3166_1": "NZ",
+    "latitude": -41.2865,
+    "longitude": 174.7762
+  },
+  {
+    "city": "Auckland",
+    "state_province": "Auckland",
+    "iso3166_2": "NZ-AUK",
+    "country": "New Zealand",
+    "iso3166_1": "NZ",
+    "latitude": -36.8485,
+    "longitude": 174.7633
+  },
+  {
+    "city": "Christchurch",
+    "state_province": "Canterbury",
+    "iso3166_2": "NZ-CAN",
+    "country": "New Zealand",
+    "iso3166_1": "NZ",
+    "latitude": -43.532,
+    "longitude": 172.6306
+  },
+  {
+    "city": "Nelson",
+    "state_province": "Nelson",
+    "iso3166_2": "NZ-NSN",
+    "country": "New Zealand",
+    "iso3166_1": "NZ",
+    "latitude": -41.2706,
+    "longitude": 173.284
+  },
+  {
+    "city": "Munich",
+    "state_province": "Bavaria",
+    "iso3166_2": "DE-BY",
+    "country": "Germany",
+    "iso3166_1": "DE",
+    "latitude": 48.1351,
+    "longitude": 11.582
+  },
+  {
+    "city": "Berlin",
+    "state_province": "Berlin",
+    "iso3166_2": "DE-BE",
+    "country": "Germany",
+    "iso3166_1": "DE",
+    "latitude": 52.52,
+    "longitude": 13.405
+  },
+  {
+    "city": "Cologne",
+    "state_province": "North Rhine-Westphalia",
+    "iso3166_2": "DE-NW",
+    "country": "Germany",
+    "iso3166_1": "DE",
+    "latitude": 50.9375,
+    "longitude": 6.9603
+  },
+  {
+    "city": "Bamberg",
+    "state_province": "Bavaria",
+    "iso3166_2": "DE-BY",
+    "country": "Germany",
+    "iso3166_1": "DE",
+    "latitude": 49.8916,
+    "longitude": 10.8916
+  },
+  {
+    "city": "Brussels",
+    "state_province": "Brussels-Capital",
+    "iso3166_2": "BE-BRU",
+    "country": "Belgium",
+    "iso3166_1": "BE",
+    "latitude": 50.8503,
+    "longitude": 4.3517
+  },
+  {
+    "city": "Antwerp",
+    "state_province": "Flanders",
+    "iso3166_2": "BE-VLG",
+    "country": "Belgium",
+    "iso3166_1": "BE",
+    "latitude": 51.2194,
+    "longitude": 4.4025
+  },
+  {
+    "city": "Bruges",
+    "state_province": "Flanders",
+    "iso3166_2": "BE-VLG",
+    "country": "Belgium",
+    "iso3166_1": "BE",
+    "latitude": 51.2093,
+    "longitude": 3.2247
+  },
+  {
+    "city": "London",
+    "state_province": "England",
+    "iso3166_2": "GB-ENG",
+    "country": "United Kingdom",
+    "iso3166_1": "GB",
+    "latitude": 51.5074,
+    "longitude": -0.1278
+  },
+  {
+    "city": "Bristol",
+    "state_province": "England",
+    "iso3166_2": "GB-ENG",
+    "country": "United Kingdom",
+    "iso3166_1": "GB",
+    "latitude": 51.4545,
+    "longitude": -2.5879
+  },
+  {
+    "city": "Edinburgh",
+    "state_province": "Scotland",
+    "iso3166_2": "GB-SCT",
+    "country": "United Kingdom",
+    "iso3166_1": "GB",
+    "latitude": 55.9533,
+    "longitude": -3.1883
+  },
+  {
+    "city": "Glasgow",
+    "state_province": "Scotland",
+    "iso3166_2": "GB-SCT",
+    "country": "United Kingdom",
+    "iso3166_1": "GB",
+    "latitude": 55.8642,
+    "longitude": -4.2518
+  },
+  {
+    "city": "Prague",
+    "state_province": "Prague",
+    "iso3166_2": "CZ-10",
+    "country": "Czechia",
+    "iso3166_1": "CZ",
+    "latitude": 50.0755,
+    "longitude": 14.4378
+  },
+  {
+    "city": "Pilsen",
+    "state_province": "Plzeň",
+    "iso3166_2": "CZ-32",
+    "country": "Czechia",
+    "iso3166_1": "CZ",
+    "latitude": 49.7384,
+    "longitude": 13.3736
+  },
+  {
+    "city": "Amsterdam",
+    "state_province": "North Holland",
+    "iso3166_2": "NL-NH",
+    "country": "Netherlands",
+    "iso3166_1": "NL",
+    "latitude": 52.3676,
+    "longitude": 4.9041
+  },
+  {
+    "city": "Copenhagen",
+    "state_province": "Capital Region",
+    "iso3166_2": "DK-84",
+    "country": "Denmark",
+    "iso3166_1": "DK",
+    "latitude": 55.6761,
+    "longitude": 12.5683
+  },
+  {
+    "city": "Warsaw",
+    "state_province": "Masovian",
+    "iso3166_2": "PL-MZ",
+    "country": "Poland",
+    "iso3166_1": "PL",
+    "latitude": 52.2297,
+    "longitude": 21.0122
+  },
+  {
+    "city": "Krakow",
+    "state_province": "Lesser Poland",
+    "iso3166_2": "PL-MA",
+    "country": "Poland",
+    "iso3166_1": "PL",
+    "latitude": 50.0647,
+    "longitude": 19.945
+  },
+  {
+    "city": "Rome",
+    "state_province": "Lazio",
+    "iso3166_2": "IT-62",
+    "country": "Italy",
+    "iso3166_1": "IT",
+    "latitude": 41.9028,
+    "longitude": 12.4964
+  },
+  {
+    "city": "Milan",
+    "state_province": "Lombardy",
+    "iso3166_2": "IT-25",
+    "country": "Italy",
+    "iso3166_1": "IT",
+    "latitude": 45.4642,
+    "longitude": 9.19
+  },
+  {
+    "city": "Barcelona",
+    "state_province": "Catalonia",
+    "iso3166_2": "ES-CT",
+    "country": "Spain",
+    "iso3166_1": "ES",
+    "latitude": 41.3851,
+    "longitude": 2.1734
+  },
+  {
+    "city": "Madrid",
+    "state_province": "Madrid",
+    "iso3166_2": "ES-MD",
+    "country": "Spain",
+    "iso3166_1": "ES",
+    "latitude": 40.4168,
+    "longitude": -3.7038
+  },
+  {
+    "city": "Paris",
+    "state_province": "Île-de-France",
+    "iso3166_2": "FR-IDF",
+    "country": "France",
+    "iso3166_1": "FR",
+    "latitude": 48.8566,
+    "longitude": 2.3522
+  },
+  {
+    "city": "Lyon",
+    "state_province": "Auvergne-Rhône-Alpes",
+    "iso3166_2": "FR-ARA",
+    "country": "France",
+    "iso3166_1": "FR",
+    "latitude": 45.764,
+    "longitude": 4.8357
+  },
+  {
+    "city": "Stockholm",
+    "state_province": "Stockholm",
+    "iso3166_2": "SE-AB",
+    "country": "Sweden",
+    "iso3166_1": "SE",
+    "latitude": 59.3293,
+    "longitude": 18.0686
+  },
+  {
+    "city": "Gothenburg",
+    "state_province": "Västra Götaland",
+    "iso3166_2": "SE-O",
+    "country": "Sweden",
+    "iso3166_1": "SE",
+    "latitude": 57.7089,
+    "longitude": 11.9746
+  },
+  {
+    "city": "Oslo",
+    "state_province": "Oslo",
+    "iso3166_2": "NO-03",
+    "country": "Norway",
+    "iso3166_1": "NO",
+    "latitude": 59.9139,
+    "longitude": 10.7522
+  },
+  {
+    "city": "Dublin",
+    "state_province": "Leinster",
+    "iso3166_2": "IE-L",
+    "country": "Ireland",
+    "iso3166_1": "IE",
+    "latitude": 53.3498,
+    "longitude": -6.2603
+  },
+  {
+    "city": "Vienna",
+    "state_province": "Vienna",
+    "iso3166_2": "AT-9",
+    "country": "Austria",
+    "iso3166_1": "AT",
+    "latitude": 48.2082,
+    "longitude": 16.3738
+  },
+  {
+    "city": "Zurich",
+    "state_province": "Zurich",
+    "iso3166_2": "CH-ZH",
+    "country": "Switzerland",
+    "iso3166_1": "CH",
+    "latitude": 47.3769,
+    "longitude": 8.5417
+  },
+  {
+    "city": "Tallinn",
+    "state_province": "Harju",
+    "iso3166_2": "EE-37",
+    "country": "Estonia",
+    "iso3166_1": "EE",
+    "latitude": 59.437,
+    "longitude": 24.7536
+  },
+  {
+    "city": "Denver",
+    "state_province": "Colorado",
+    "iso3166_2": "US-CO",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 39.7392,
+    "longitude": -104.9903
+  },
+  {
+    "city": "Portland",
+    "state_province": "Oregon",
+    "iso3166_2": "US-OR",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 45.5152,
+    "longitude": -122.6784
+  },
+  {
+    "city": "San Diego",
+    "state_province": "California",
+    "iso3166_2": "US-CA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 32.7157,
+    "longitude": -117.1611
+  },
+  {
+    "city": "Asheville",
+    "state_province": "North Carolina",
+    "iso3166_2": "US-NC",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 35.5951,
+    "longitude": -82.5515
+  },
+  {
+    "city": "Grand Rapids",
+    "state_province": "Michigan",
+    "iso3166_2": "US-MI",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 42.9634,
+    "longitude": -85.6681
+  },
+  {
+    "city": "Chicago",
+    "state_province": "Illinois",
+    "iso3166_2": "US-IL",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 41.8781,
+    "longitude": -87.6298
+  },
+  {
+    "city": "Seattle",
+    "state_province": "Washington",
+    "iso3166_2": "US-WA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 47.6062,
+    "longitude": -122.3321
+  },
+  {
+    "city": "Austin",
+    "state_province": "Texas",
+    "iso3166_2": "US-TX",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 30.2672,
+    "longitude": -97.7431
+  },
+  {
+    "city": "Boston",
+    "state_province": "Massachusetts",
+    "iso3166_2": "US-MA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 42.3601,
+    "longitude": -71.0589
+  },
+  {
+    "city": "Philadelphia",
+    "state_province": "Pennsylvania",
+    "iso3166_2": "US-PA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 39.9526,
+    "longitude": -75.1652
+  },
+  {
+    "city": "Brooklyn",
+    "state_province": "New York",
+    "iso3166_2": "US-NY",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 40.6782,
+    "longitude": -73.9442
+  },
+  {
+    "city": "Milwaukee",
+    "state_province": "Wisconsin",
+    "iso3166_2": "US-WI",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 43.0389,
+    "longitude": -87.9065
+  },
+  {
+    "city": "Richmond",
+    "state_province": "Virginia",
+    "iso3166_2": "US-VA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 37.5407,
+    "longitude": -77.436
+  },
+  {
+    "city": "Cincinnati",
+    "state_province": "Ohio",
+    "iso3166_2": "US-OH",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 39.1031,
+    "longitude": -84.512
+  },
+  {
+    "city": "St. Louis",
+    "state_province": "Missouri",
+    "iso3166_2": "US-MO",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 38.627,
+    "longitude": -90.1994
+  },
+  {
+    "city": "Tampa",
+    "state_province": "Florida",
+    "iso3166_2": "US-FL",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 27.9506,
+    "longitude": -82.4572
+  },
+  {
+    "city": "Minneapolis",
+    "state_province": "Minnesota",
+    "iso3166_2": "US-MN",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 44.9778,
+    "longitude": -93.265
+  },
+  {
+    "city": "Burlington",
+    "state_province": "Vermont",
+    "iso3166_2": "US-VT",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 44.4759,
+    "longitude": -73.2121
+  },
+  {
+    "city": "Portland",
+    "state_province": "Maine",
+    "iso3166_2": "US-ME",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 43.6591,
+    "longitude": -70.2568
+  },
+  {
+    "city": "Atlanta",
+    "state_province": "Georgia",
+    "iso3166_2": "US-GA",
+    "country": "United States",
+    "iso3166_1": "US",
+    "latitude": 33.749,
+    "longitude": -84.388
+  },
+  {
+    "city": "Toronto",
+    "state_province": "Ontario",
+    "iso3166_2": "CA-ON",
+    "country": "Canada",
+    "iso3166_1": "CA",
+    "latitude": 43.651,
+    "longitude": -79.347
+  },
+  {
+    "city": "Vancouver",
+    "state_province": "British Columbia",
+    "iso3166_2": "CA-BC",
+    "country": "Canada",
+    "iso3166_1": "CA",
+    "latitude": 49.2827,
+    "longitude": -123.1207
+  },
+  {
+    "city": "Montreal",
+    "state_province": "Quebec",
+    "iso3166_2": "CA-QC",
+    "country": "Canada",
+    "iso3166_1": "CA",
+    "latitude": 45.5017,
+    "longitude": -73.5673
+  },
+  {
+    "city": "Calgary",
+    "state_province": "Alberta",
+    "iso3166_2": "CA-AB",
+    "country": "Canada",
+    "iso3166_1": "CA",
+    "latitude": 51.0447,
+    "longitude": -114.0719
+  },
+  {
+    "city": "Halifax",
+    "state_province": "Nova Scotia",
+    "iso3166_2": "CA-NS",
+    "country": "Canada",
+    "iso3166_1": "CA",
+    "latitude": 44.6488,
+    "longitude": -63.5752
+  },
+  {
+    "city": "Mexico City",
+    "state_province": "Mexico City",
+    "iso3166_2": "MX-CMX",
+    "country": "Mexico",
+    "iso3166_1": "MX",
+    "latitude": 19.4326,
+    "longitude": -99.1332
+  },
+  {
+    "city": "Tijuana",
+    "state_province": "Baja California",
+    "iso3166_2": "MX-BCN",
+    "country": "Mexico",
+    "iso3166_1": "MX",
+    "latitude": 32.5149,
+    "longitude": -117.0382
+  },
+  {
+    "city": "Monterrey",
+    "state_province": "Nuevo León",
+    "iso3166_2": "MX-NLE",
+    "country": "Mexico",
+    "iso3166_1": "MX",
+    "latitude": 25.6866,
+    "longitude": -100.3161
+  },
+  {
+    "city": "Guadalajara",
+    "state_province": "Jalisco",
+    "iso3166_2": "MX-JAL",
+    "country": "Mexico",
+    "iso3166_1": "MX",
+    "latitude": 20.6597,
+    "longitude": -103.3496
+  },
+  {
+    "city": "Ensenada",
+    "state_province": "Baja California",
+    "iso3166_2": "MX-BCN",
+    "country": "Mexico",
+    "iso3166_1": "MX",
+    "latitude": 31.8667,
+    "longitude": -116.5964
+  }
+]

pipeline/prompts/brewery_system_prompt.txt

@@ -0,0 +1,425 @@
+================================================================================
+BREWERY DATA GENERATION - COMPREHENSIVE SYSTEM PROMPT
+================================================================================
+
+ROLE AND OBJECTIVE
+You are an experienced brewmaster and owner of a local craft brewery. Your task
+is to create a distinctive, authentic name and a detailed description for your
+brewery that genuinely reflects your specific location, your brewing philosophy,
+the local culture, and your connection to the community.
+
+The brewery must feel real and grounded in its specific place—not generic or
+interchangeable with breweries from other regions. Every detail should build
+authenticity and distinctiveness.
+
+================================================================================
+FORBIDDEN PHRASES AND CLICHÉS
+================================================================================
+
+NEVER USE THESE OVERUSED CONSTRUCTIONS (even in modified form):
+- "Love letter to" / "tribute to" / "ode to"
+- "Rolling hills" / "picturesque landscape" / "scenic beauty"
+- "Every sip tells a story" / "every pint tells a story" / "transporting you"
+- "Come for X, stay for Y" formula (Come for beer, stay for...)
+- "Rich history/traditions" / "storied past" / "storied brewing tradition"
+- "Passion" as a generic descriptor ("crafted with passion", "our passion")
+- "Woven into the fabric" / "echoes of" / "steeped in"
+- "Ancient roots" / "timeless traditions" / "time-honored heritage"
+- Opening ONLY with landscape/geography (no standalone "Nestled...", "Where...")
+- "Where tradition meets innovation"
+- "Celebrating the spirit of [place]"
+- "Raised on the values of" / "rooted in the values of"
+- "Taste of [place]" / "essence of [place]"
+- "From our family to yours"
+- "Brewing excellence" / "committed to excellence"
+- "Bringing people together" (without showing HOW)
+- "Honoring local heritage" (without specifics)
+
+================================================================================
+SEVEN OPENING APPROACHES - ROTATE BETWEEN THESE
+================================================================================
+
+1. BEER STYLE ORIGIN ANGLE
+   Start by identifying a specific beer style historically made in or
+   influenced by the region. Explain why THIS place inspired that style.
+   Example Foundation: "Belgian Trappist ales developed from monastic traditions
+   in the Ardennes; our brewery continues that contemplative approach..."
+
+2. BREWING CHALLENGE / ADVANTAGE ANGLE
+   Begin with a specific environmental or geographic challenge that shapes
+   the brewery's approach. Water hardness, altitude, climate, ingredient scarcity.
+   Example Foundation: "High-altitude fermentation requires patience; at 1,500m,
+   our lagers need 8 weeks to develop the crisp finish..."
+
+3. FOUNDING STORY / PERSONAL MOTIVATION
+   Open with why the founder started THIS brewery HERE. Personal history,
+   escape from corporate work, multi-generational family legacy, career change.
+   Example Foundation: "After 20 years in finance, I returned to my hometown to
+   revive my grandfather's closed brewery using his original recipe notes..."
+
+4. SPECIFIC LOCAL INGREDIENT / RESOURCE
+   Lead with a unique input source: special water, rare hops grown locally,
+   grain from a specific mill, honey from local apiaries, barrel aging with
+   local wood.
+   Example Foundation: "The cold springs below Sniffels Peak provide water so soft
+   it inspired our signature pale lager..."
+
+5. CONTRADICTION / UNEXPECTED ANGLE
+   Start with a surprising fact about the place that defies stereotype.
+   Example Foundation: "Nobody expects beer culture in a Muslim-majority city,
+   yet our secular neighborhood has deep roots in 1920s beer halls..."
+
+6. LOCAL EVENT / CULTURAL MOMENT
+   Begin with a specific historical moment, festival, cultural practice, or
+   seasonal tradition in the place.
+   Example Foundation: "Every October, the hop harvest brings itinerant workers
+   and tradition. Our brewery grew from a harvest celebration in 2008..."
+
+7. TANGIBLE PHYSICAL DETAIL
+   Open by describing a concrete architectural or geographic feature: building
+   age, material, location relative to notable structures, layout, history of
+   the space.
+   Example Foundation: "This 1887 mill house once crushed grain; the original
+   water wheel still runs below our fermentation room..."
+
+================================================================================
+SPECIFICITY AND CONCRETENESS REQUIREMENTS
+================================================================================
+
+DO NOT GENERALIZE. Every brewery description must include:
+
+✓ At least ONE concrete proper noun or specific reference:
+  - Actual local landmarks (mountain name, river name, street, neighborhood)
+  - Specific business partner or supplier name (if real to the region)
+  - Named local cultural event or historical period
+  - Specific beer style(s) with regional significance
+  - Actual geographic feature (e.g., "the volcanic ash in our soil")
+
+✓ Mention specific beer styles relevant to the region's culture:
+  - German Bavaria: Dunkelweizen, Märzen, Kellerbier, Helles
+  - Belgian/Flemish: Lambic, Trappist, Strong Dark Ale
+  - British Isles: Brown Ale, Real Ale, Bitter, Cask Ale
+  - Czech: Pilsner, Bohemian Lager
+  - IPA/Hoppy: American regions, UK (origin)
+  - New Zealand/Australia: Hop-forward, experimental
+  - Japanese: Clean lagers, sake influence
+  - Mexican: Lager-centric, sometimes citrus
+
+✓ Name concrete brewing challenges or advantages:
+  Examples: water minerality, altitude, temperature swings, grain varieties,
+  humidity, wild yeasts in the region, traditional equipment preserved in place
+
+✓ Use sensory language SPECIFIC to the place:
+  NOT: "beautiful views"  →  "the copper beech trees turn rust-colored by
+                              September"
+  NOT: "charming"        →  "the original tile floor from 1924 still mosaic-patterns
+                              the taproom"
+  NOT: "authentic"       →  "the water chiller uses the original 1950s ammonia system"
+
+✓ Avoid describing multiple regions with the same adjectives:
+  Don't say every brewery is "cozy" or "vibrant" or "historic"—be specific
+  about WHAT makes this one different from others in different regions.
+
+================================================================================
+STRUCTURAL PATTERNS - MIX THESE UP
+================================================================================
+
+NOT every description should follow: legacy → current brewing → call to action
+
+TEMPLATE ROTATION (these are EXAMPLES, not formulas):
+
+TEMPLATE A: [Region origin] → [specific challenge] → [how we adapted] → [result]
+  "The Saône River flooded predictably each spring. Medieval brewers learned
+  to schedule production around it. We use the same seasonal rhythm..."
+
+TEMPLATE B: [Ingredient story] → [technique developed because of it] → [distinctive result]
+  "Our barley terraces face southwest; the afternoon sun dries the crop weeks
+  before northern valleys. This inspired our crisp, mineral-forward pale ale..."
+
+TEMPLATE C: [Personal/family history (without generic framing)] → [specific challenge overcome] → [philosophy]
+  "My mother was a chemist studying water quality; she noticed the local supply
+  had unusual pH. Rather than fight it, we formulated our entire range around
+  it. The sulfate content sharpens our bitters..."
+
+TEMPLATE D: [Describe the physical space in detail] → [how space enables brewing style] → [sensory experience]
+  "The brewhouse occupies a converted 1960s chemical factory. The stainless steel
+  vats still bear faded original markings. The building's thermal mass keeps
+  fermentation stable without modern refrigeration..."
+
+TEMPLATE E: [Unexpected contradiction] → [explanation] → [brewing philosophy]
+  "In a region famous for wine, we're a beer-only operation. We embrace that
+  outsider status and brew adventurously, avoiding the 'respect tradition'
+  pressure wine makes locals feel..."
+
+TEMPLATE F: [Community role, specific] → [what that demands] → [brewing expression]
+  "We're the only gathering space in the village that stays open after 10pm.
+  That responsibility means brewing beers that pair with conversation, not
+  provocation. Sessionable, food-friendly, endlessly drinkable..."
+
+TEMPLATE G: [Backward chronology] → [how practices persist] → [what's evolved]
+  "Our great-grandfather hand-packed bottles in 1952. We still own his bench.
+  Even though we use machines now, the pace he set—careful, thoughtful—shapes
+  every decision. Nothing about us is fast..."
+
+SOMETIMES skip the narrative entirely and just describe:
+  "We brew four core beers—a dry lager, a copper ale, a wheat beer, and a hop-
+  forward pale. The range itself tells our story: accessible, varied,
+  unpretentious. No flagship. No hero beer. Balance."
+
+================================================================================
+REGIONAL AUTHENTICITY GUIDELINES
+================================================================================
+
+GERMAN / ALPINE / CENTRAL EUROPEAN
+- Discuss water hardness and mineral content
+- Reference specific beer laws (Reinheitsgebot, Bavarian purity traditions)
+- Name specific styles: Kellerbier, Märzen, Dunkelweizen, Helles, Alt, Zwickel
+- Mention lager fermentation dominance and cool-cave advantages
+- Consider beer hall culture, tradition of communal spaces
+- Discuss barrel aging if applicable
+- Reference precision/engineering in brewing approach
+- Don't romanticize; emphasis can be on technique and consistency
+
+MEDITERRANEAN / SOUTHERN EUROPEAN
+- Reference local wine culture (compare or contrast with brewing)
+- Mention grape varieties if relevant (some regions have wine-brewery overlap)
+- Discuss sun exposure, heat challenges during fermentation
+- Ingredient sourcing: local herbs, citrus, wheat quality
+- May emphasize Mediterranean sociability and gathering spaces
+- Consider how northern European brewing tradition transplanted here
+- Water source and quality specific to region
+- Seasonal agricultural connections (harvest timing, etc.)
+
+ANGLO-SAXON / BRITISH ISLES / SCANDINAVIAN
+- Real ale, cask conditioning, hand-pulled pints
+- IPA heritage (if British, England specifically; if American, different innovation story)
+- Hops: specific varietal heritage (Fuggle, Golding, Cascade, etc.)
+- Pub culture and community gathering
+- Ales: top-fermented, warmer fermentation temperatures
+- May emphasize working-class history or rural traditions
+- Cider/mead/fermented heritage alongside beer
+
+NEW WORLD (US, AUSTRALIA, NZ, SOUTH AFRICA)
+- Emphasize experimentation and lack of brewing "rules"
+- Ingredient sourcing: local grain growers, foraged hops, local suppliers
+- May reference mining heritage, recent settlement, diverse immigration
+- Craft beer boom influence: how does this brewery differentiate?
+- Often: bold flavors, high ABVs, creative adjuncts
+- Can emphasize anti-tradition or deliberate rule-breaking
+- Emphasis on farmer partnerships and local food scenes
+
+SMALL VILLAGES / RURAL AREAS
+- Brewery likely serves as actual gathering place—explain HOW
+- Ingredient sourcing highly local (grain from X farm, water from Y spring)
+- May be family operation or multi-generation story
+- Role in community identity and events
+- Accessibility and lack of pretension
+- Seasonal rhythm and agricultural calendar influence
+- Risk: Don't make it overly quaint or "simpler times" nostalgic
+
+URBAN / NEIGHBORHOOD-BASED
+- Distinctive neighborhood identity (don't just say "vibrant")
+- Specific business community or residential character
+- Street-level visibility and casual drop-in culture
+- May emphasize diversity, immigrant heritage, gentrification navigation
+- Smaller brewing scale in dense area (space constraints)
+- Walking-distance customer base instead of destination draw
+- May have stronger food pairing focus (food truck culture, restaurant neighbors)
+
+WINE REGIONS (Italy, France, Spain, Germany's Mosel, etc.)
+- Show awareness of wine's prestige locally
+- Explain why brewing exists here despite wine dominance
+- Does brewery respect wine or deliberately provide alternative?
+- Ingredient differences: water quality suited to beer, not wine
+- Brewing approach: precise, clean—influenced by wine mentality
+- May emphasize beer's sociability vs. wine's formality
+- Historical context: beer predates or coexists with wine tradition
+
+BEER-HERITAGE HOTSPOTS (Belgium, Germany, UK, Czech Republic)
+- Can't ignore the weight of history without acknowledging it
+- Do you innovate within tradition or break from it? Say which.
+- Specific pride in one style over others (Lambic specialist, Trappist-inspired, etc.)
+- May emphasize family legacy or generational knowledge
+- Regional identity VERY strong—brewery reflects this unapologetically
+- Risk: Avoid claiming to "honor" or "continue" without specifics
+
+================================================================================
+TONE VARIATIONS - NOT ALL BREWERIES ARE SOULFUL
+================================================================================
+
+These descriptions should NOT all sound romantic, quaint, or emotionally
+passionate. These are alternative tones:
+
+IRREVERENT / HUMOROUS
+  "We're brewing beer because wine required too much prayer. Less spirituality,
+   more hops. Our ales are big, unpolished, and perfect after a day's work."
+
+MATTER-OF-FACT / ENGINEERING-FOCUSED
+  "Brewing is chemistry. We source ingredient components, control variables,
+   and optimize for reproducibility. If that sounds clinical, good—consistency
+   is our craft."
+
+PROUDLY UNPRETENTIOUS / WORKING-CLASS
+  "This isn't farm-to-table aspirational nonsense. It's a neighborhood beer.
+   $4 pints. No reservations. No sipping notes. Tastes good, fills the glass,
+   keeps you coming back."
+
+MINIMALIST / DIRECT
+  "We brew three beers. They're good. Come drink one."
+
+BUSINESS-FOCUSED / PRACTICAL
+  "Starting a brewery in 2015 meant finding a niche. We're the only nano-
+   brewery serving the airport district. Our rapid turnover and distribution
+   focus differentiate us from weekend hobbyists."
+
+CONFRONTATIONAL / REBELLIOUS
+  "Craft beer got boring. Expensive IPAs and flavor-chasing. We're brewing
+   wheat beers and forgotten styles because fashion is temporary; good beer is timeless."
+
+MIX these tones across your descriptions. Some breweries should sound romantic
+and place-proud. Others should sound irreverent or practical.
+
+================================================================================
+NARRATIVE CLICHÉS TO ABSOLUTELY AVOID
+================================================================================
+
+1. THE "HIDDEN GEM" FRAMING
+   Don't use discovery language: "hidden," "lesser-known," "off the beaten path,"
+   "tucked away." Implies marketing speak, not authenticity.
+
+2. OVERT NOSTALGIA / "SIMPLER TIMES"
+   Don't appeal to vague sense that past was better: "yearning for," "those
+   days," "how things used to be." Lazy and off-putting.
+
+3. EMPTY "GATHERING PLACE" CLAIMS
+   Don't just assert "we bring people together." Show HOW: local workers' lunch
+   spot? Trivia night tradition? Live music venue? Political meeting ground?
+
+4. "SPECIAL" WITHOUT EVIDENCE
+   Don't declare location is "special" or "unique." SHOW what makes it distinct
+   through specific details, not assertion.
+
+5. "WE BELIEVE IN" AS PLACEHOLDER
+   Every brewery claims to "believe in" quality, community, craft, sustainability.
+   These are empty. What specific belief drives THIS brewery's choices?
+
+6. "ESCAPE / RETREAT" FRAMING
+   Don't suggest beer allows people to escape reality, retreat from the world,
+   or "get away." Implies you don't trust the place itself to be compelling.
+
+7. SUPERLATIVE CLAIMS
+   Don't use: "finest," "best," "most authentic," "truly legendary." Let details
+   prove these implied claims instead.
+
+8. PASSIVE VOICE ABOUT YOUR OWN BREWERY
+   Avoid: "beloved by locals," "known for its," "celebrated for." Active voice:
+   what does the brewery actively DO?
+
+================================================================================
+LENGTH AND CONTENT REQUIREMENTS
+================================================================================
+
+TARGET LENGTH: 120-180 words
+- Long enough to establish place and brewing philosophy
+- Short enough to avoid meandering or repetition
+- Specific enough that brewery feels real and unreplicable
+
+REQUIRED ELEMENTS (at least ONE each):
+✓ Concrete location reference (proper noun, landmark, geographic feature)
+✓ One specific brewing detail (challenge, advantage, technique, ingredient)
+✓ Sensory language specific to the place (NOT generic adjectives)
+✓ Distinct tone/voice (don't all sound the same quiet reverence)
+
+OPTIONAL ELEMENTS:
+- Name 1-2 specific beer styles or beer names
+- Personal/family story (if it illuminates why brewery exists here)
+- Ingredient sourcing or supply chain detail
+- Community role (with evidence, not assertion)
+- Regional historical context (brief, specific)
+
+WORD ECONOMY:
+- Don't waste words on "we believe in quality" or "committed to excellence"
+- Don't use filler adjectives: "authentic," "genuine," "real," "true," "local"
+  (these should be IMPLIED by specific details)
+- Every sentence should add information, flavor, or distinctive detail
+
+================================================================================
+SENSORY LANGUAGE GUIDELINES
+================================================================================
+
+AVOID THESE GENERIC SENSORY WORDS (they're lazy placeholders):
+- "Beautiful," "picturesque," "gorgeous," "stunning"
+- "Warm," "cozy," "inviting" (without context)
+- "Vibrant," "lively," "energetic" (without examples)
+- "Charming," "quaint," "rustic" (without specifics)
+
+USE INSTEAD: Specific, concrete sensory details
+- Colors: "copper beech," "rust-stained brick," "frost-blue shutters"
+- Textures: "the grain of wooden barrel hoops," "hand-smoothed stone," "grime-darkened windows"
+- Sounds: "the hiss of the hand-pump," "coin-drop in the old register," "church bells on Sunday"
+- Smells: "yeast-heavy floor," "wet limestone," "Hallertau hop resin"
+- Tastes: (in the beer) "mineral-sharp," "sulfate clarity," "heather honey notes"
+
+EXAMPLE SENSORY COMPARISON:
+AVOID: "Our brewery captures the essence of the region's rustic charm."
+USE:   "The five-meter stone walls keep fermentation at 12°C without refrigeration.
+        On warm days, water drips from moss-covered blocks—the original cooling
+        system that hasn't changed in 150 years."
+
+================================================================================
+DIVERSITY ACROSS DATASET - WHAT NOT TO REPEAT
+================================================================================
+
+Since you're generating many breweries, ensure variety by:
+
+□ Alternating tone (soulful → irreverent → matter-of-fact → working-class, etc.)
+□ Varying opening approach (don't use beer-style origin twice in a row)
+□ Different geographic contexts (don't make all small villages sound the same)
+□ Distinct brewery sizes/models (nano-brewery, family operation, investor-backed, etc.)
+□ Various types of "draw" (neighborhood destination vs. local-only vs. tourist
+  attraction vs. untouched community staple)
+□ Diverse relationship to beer history/tradition (embrace it, subvert it, ignore it)
+□ Different community roles (political space, athlete hangout, food destination,
+  working person's bar, experimentation lab, etc.)
+
+If you notice yourself using the same phrasing twice within three breweries,
+STOP and take a completely different approach for the next one.
+
+================================================================================
+QUALITY CHECKLIST
+================================================================================
+
+Before submitting your brewery description, verify:
+
+□ Zero clichés from the FORBIDDEN list appear anywhere
+□ At least one specific proper noun or concrete reference included
+□ No more than two generic adjectives in the entire description
+□ The brewery is genuinely unreplicable (wouldn't work in a different location)
+□ Tone matches a SPECIFIC angle (not generic reverence)
+□ Opening sentence is distinctive and unexpected
+□ No sentence says the same thing twice in different words
+□ At least one detail is surprising or specific to this place
+□ The description would make sense ONLY for this location/region
+□ "Passion," "tradition," "community" either don't appear or appear with
+  specific context/evidence
+
+================================================================================
+OUTPUT FORMAT
+================================================================================
+
+Return ONLY a valid JSON object with exactly two keys:
+{
+  "name": "Brewery Name Here",
+  "description": "Full description text here..."
+}
+
+Requirements:
+- name: 2-5 words, distinctive, memorable
+- description: 120-180 words, follows all guidelines above
+- Valid JSON (escaped quotes, no line breaks in strings)
+- No markdown, no backticks, no code formatting
+- No preamble before the JSON
+- No trailing text after the JSON
+- No explanations or commentary
+
+================================================================================

pipeline/prompts/brewery_system_prompt_expanded.txt

@@ -0,0 +1,200 @@
+================================================================================
+BREWERY DATA GENERATION SYSTEM PROMPT
+
+ROLE AND OBJECTIVE
+You are an experienced brewmaster creating brewery descriptions grounded in the
+given city and country. The writing must feel specific, plausible, and local
+without sounding formulaic or repetitive.
+
+Primary goal: produce varied outputs across many cities in one run.
+Do NOT use the same template repeatedly.
+
+================================================================================
+ANTI-REPETITION RULES (CRITICAL)
+
+Avoid recurring boilerplate patterns. Especially avoid repeatedly using:
+
+- "The soft spring water beneath..."
+- fixed mineral ppm patterns in every entry
+- "1930s copper still/mash tun" in every entry
+- "the air smells of..." in every entry
+- "No stainless steel" / anti-modernization comparison
+- year-heavy historical stacking in every paragraph
+
+For each brewery, choose a DIFFERENT primary lens from this set:
+
+1) Local ingredient chain
+2) Fermentation/process decision
+3) Building/space constraint
+4) Workforce/customer culture
+5) Regional beer tradition adapted locally
+6) Climate/seasonality challenge
+
+Use only one primary lens plus one supporting detail.
+Do not combine all lenses every time.
+
+Vary rhythm and structure:
+
+- Some descriptions should be concise and direct.
+- Some can be narrative.
+- Some can be technical.
+- Do not start more than 2 descriptions in a row with the same sentence shape.
+
+================================================================================
+FORBIDDEN PHRASES
+
+NEVER USE THESE (even in modified form):
+
+"Love letter to" / "tribute to" / "ode to" / "rolling hills" / "picturesque"
+
+"Every sip tells a story" / "Come for X, stay for Y" / "Where tradition meets innovation"
+
+"Rich history" / "ancient roots" / "timeless traditions" / "time-honored heritage"
+
+"Passion" (standalone descriptor) / "brewing excellence" / "commitment to quality"
+
+"Authentic" / "genuine" / "real" / "true" (SHOW these, don't state them)
+
+"Bringing people together" (without HOW) / "community gathering place" (without proof)
+
+"Hidden gem" / "secret" / "lesser-known" / "beloved by locals"
+
+Generic adjectives: "beautiful," "gorgeous," "lovely," "cozy," "charming," "vibrant"
+
+Vague temporal claims: "simpler times," "the good old days," "escape from the modern world"
+
+Passive voice: "is known for," "has become famous for," "has earned a reputation"
+
+================================================================================
+OPENING APPROACHES (Choose ONE)
+
+BEER STYLE ORIGIN: Start with a specific historical beer style from this
+region, explain why this place created it, show how your brewery continues it.
+Key: style + local reason + current execution
+
+BREWING CHALLENGE: Begin with a specific environmental constraint (altitude,
+water hardness, temperature, endemic yeasts). Explain the technical consequence
+and what decision you made because of it.
+Key: constraint + consequence + response
+
+FOUNDING STORY: Why did the founder return/move HERE? What did they discover?
+What specific brewing decision followed? Include a concrete artifact (logs, equipment).
+Key: motivation + discovery + decision
+
+LOCAL INGREDIENT: What unique resource defines your brewery? Why is it unique?
+What brewing constraint or opportunity does it create?
+Key: ingredient + locality + process effect
+
+CONTRADICTION: What is the region famous for? Why does your brewery do the
+opposite? Make the contradiction a strength, not an apology.
+Key: regional norm + divergence + result
+
+CULTURAL MOMENT: What specific seasonal tradition or event shapes your brewery?
+How do you connect to it? What brewing decisions follow?
+Key: event + relationship + brewing choice
+
+PHYSICAL SPACE: Describe a specific architectural feature with date/material.
+How does it create technical advantage? What sensory details matter? Why keep
+constraints instead of modernizing?
+Key: feature + consequence + sensory note
+
+================================================================================
+SPECIFICITY REQUIREMENTS
+
+Every brewery description MUST include:
+
+CONCRETE PROPER NOUNS (at least 2)
+
+Named geographic features relevant to the prompt location.
+
+Named local suppliers or historical events specific to the region.
+
+BREWING DETAIL (exactly 1-2)
+
+Examples: mash schedule choice, fermentation temperature strategy,
+ingredient handling, yeast management, packaging decision.
+
+Numeric values are OPTIONAL.
+Only use numbers when highly plausible.
+Do not force ppm chemistry in every description.
+
+Avoid making up overly specific historical claims unless they are broadly plausible.
+
+SENSORY DETAIL (at least 1)
+Must be local and concrete (sound/smell/texture/visual).
+Do not reuse identical sensory phrasing across outputs.
+
+PROOF TEST
+Could this description be pasted onto another city unchanged?
+If yes, make it more local.
+
+If no, proceed.
+
+================================================================================
+TONE VARIATIONS
+
+Rotate tones consciously.
+
+Do not lock into one tone for all cities. Choose one per city.
+
+IRREVERENT: blunt, anti-hype, practical.
+
+MATTER-OF-FACT: technical and concise.
+
+WORKING-CLASS PROUD: utility, affordability, regulars.
+
+MINIMALIST: short, sparse, direct.
+
+NOSTALGIC-GROUNDED: legacy through tangible artifacts.
+
+================================================================================
+LENGTH & CONTENT REQUIREMENTS
+
+TARGET LENGTH: 90-170 words
+
+REQUIRED ELEMENTS:
+
+At least 2 concrete proper nouns
+
+At least 1 brewing-specific detail
+
+At least 1 local sensory detail
+
+Consistent tone throughout (irreverent, matter-of-fact, working-class, nostalgic, etc.)
+
+One distinctive detail that proves the brewery could ONLY exist in this location
+
+DO NOT INCLUDE:
+
+Generic adjectives without evidence: "authentic," "genuine," "soulful," "passionate"
+
+Vague community claims without HOW: "gathering place," "beloved," "where people come together"
+
+Marketing language: "award-winning," "nationally recognized," "craft quality"
+
+Fillers: "and more," "creating memories," "for all to enjoy"
+
+Predictions: "we're working on," "coming soon," "we plan to"
+
+Do not repeat the same structural motifs across outputs in one batch.
+
+================================================================================
+OUTPUT FORMAT
+
+Return ONLY a valid JSON object with exactly two keys:
+{
+"name": "Brewery Name Here",
+"description": "Full description text here..."
+}
+
+Requirements:
+
+name: 2-5 words, distinctive, memorable
+
+description: 90-170 words, follows all guidelines
+
+Valid JSON (properly escaped quotes, no line breaks)
+
+No markdown, backticks, or code formatting
+
+No preamble or trailing text after JSON

pipeline/src/biergarten_data_generator.cpp

@@ -0,0 +1,162 @@
+#include "biergarten_data_generator.h"
+
+#include <spdlog/spdlog.h>
+
+#include <algorithm>
+#include <filesystem>
+#include <future>
+#include <iterator>
+#include <random>
+
+#include "data_generation/llama_generator.h"
+#include "data_generation/mock_generator.h"
+#include "json_handling/json_loader.h"
+#include "wikipedia/wikipedia_service.h"
+
+BiergartenDataGenerator::BiergartenDataGenerator(
+   const ApplicationOptions& options, std::shared_ptr<WebClient> web_client)
+   : options_(options), webClient_(std::move(web_client)) {}
+
+auto BiergartenDataGenerator::InitializeGenerator()
+   -> std::unique_ptr<DataGenerator> {
+   spdlog::info("Initializing brewery generator...");
+
+   std::unique_ptr<DataGenerator> generator;
+   if (options_.model_path.empty()) {
+      generator = std::make_unique<MockGenerator>();
+      spdlog::info("[Generator] Using MockGenerator (no model path provided)");
+   } else {
+      auto llama_generator = std::make_unique<LlamaGenerator>();
+      llama_generator->SetSamplingOptions(options_.temperature, options_.top_p,
+                                          options_.seed);
+      llama_generator->SetContextSize(options_.n_ctx);
+      spdlog::info(
+          "[Generator] Using LlamaGenerator: {} (temperature={}, top-p={}, "
+          "n_ctx={}, seed={})",
+          options_.model_path, options_.temperature, options_.top_p,
+          options_.n_ctx, options_.seed);
+      generator = std::move(llama_generator);
+   }
+   generator->Load(options_.model_path);
+
+   return generator;
+}
+
+auto BiergartenDataGenerator::QueryCitiesWithCountries()
+   -> std::vector<Location> {
+   spdlog::info("\n=== GEOGRAPHIC DATA OVERVIEW ===");
+
+   std::filesystem::path locations_path = "locations.json";
+   if (!std::filesystem::exists(locations_path)) {
+      const std::filesystem::path cache_path =
+          std::filesystem::path(options_.cache_dir) / "locations.json";
+      if (std::filesystem::exists(cache_path)) {
+         locations_path = cache_path;
+      }
+   }
+
+   auto all_locations = JsonLoader::LoadLocations(locations_path.string());
+   spdlog::info("  Locations available: {}", all_locations.size());
+
+   const size_t sample_count = std::min<size_t>(4, all_locations.size());
+   std::vector<Location> sampled_locations;
+   sampled_locations.reserve(sample_count);
+
+   std::random_device random_generator;
+   std::sample(all_locations.begin(), all_locations.end(),
+               std::back_inserter(sampled_locations), sample_count,
+               random_generator);
+
+   spdlog::info("  Sampled locations: {}", sampled_locations.size());
+   return sampled_locations;
+}
+
+auto BiergartenDataGenerator::EnrichWithWikipedia(
+   const std::vector<Location>& cities) -> std::vector<EnrichedCity> {
+   std::vector<EnrichedCity> enriched;
+   enriched.reserve(cities.size());
+
+   std::vector<std::future<EnrichedCity>> pending;
+   pending.reserve(cities.size());
+
+   for (const auto& city : cities) {
+      pending.push_back(std::async(std::launch::async,
+                                   [web_client = webClient_, city]() {
+                                      WikipediaService wikipedia_service(
+                                          web_client);
+                                      const std::string region_context =
+                                          wikipedia_service.GetSummary(
+                                              city.city, city.country);
+                                      spdlog::debug(
+                                          "[Pipeline] Region context for {}: {}",
+                                          city.city, region_context);
+                                      return EnrichedCity{city, region_context};
+                                   }));
+   }
+
+   for (auto& task : pending) {
+      enriched.push_back(task.get());
+   }
+
+   return enriched;
+}
+
+void BiergartenDataGenerator::GenerateBreweries(
+    DataGenerator& generator, const std::vector<EnrichedCity>& cities) {
+   spdlog::info("\n=== SAMPLE BREWERY GENERATION ===");
+   generatedBreweries_.clear();
+
+   size_t skipped_count = 0;
+
+   for (const auto& enriched_city : cities) {
+      try {
+         auto brewery = generator.GenerateBrewery(enriched_city.location.city,
+                                                  enriched_city.location.country,
+                                                  enriched_city.region_context);
+         generatedBreweries_.push_back({enriched_city.location, brewery});
+      } catch (const std::exception& e) {
+         ++skipped_count;
+         spdlog::warn(
+             "[Pipeline] Skipping city '{}' ({}): brewery generation failed: {}",
+             enriched_city.location.city, enriched_city.location.country,
+             e.what());
+      }
+   }
+
+   if (skipped_count > 0) {
+      spdlog::warn("[Pipeline] Skipped {} city/cities due to generation "
+                   "errors",
+                   skipped_count);
+   }
+}
+
+void BiergartenDataGenerator::LogResults() const {
+   spdlog::info("\n=== GENERATED DATA DUMP ===");
+   size_t index = 1;
+   for (const auto& entry : generatedBreweries_) {
+      spdlog::info("{}. city=\"{}\" country=\"{}\" state=\"{}\" "
+                   "iso3166_2={} lat={} lon={}",
+                   index, entry.location.city, entry.location.country,
+                   entry.location.state_province, entry.location.iso3166_2,
+                   entry.location.latitude, entry.location.longitude);
+      spdlog::info("   brewery_name=\"{}\"", entry.brewery.name);
+      spdlog::info("   brewery_description=\"{}\"", entry.brewery.description);
+      ++index;
+   }
+}
+
+auto BiergartenDataGenerator::Run() -> int {
+   try {
+      auto generator = InitializeGenerator();
+      auto cities = QueryCitiesWithCountries();
+      auto enriched = EnrichWithWikipedia(cities);
+      GenerateBreweries(*generator, enriched);
+      LogResults();
+
+      spdlog::info("\nOK: Pipeline completed successfully");
+      return 0;
+   } catch (const std::exception& e) {
+      spdlog::error("ERROR: Pipeline failed: {}", e.what());
+      return 1;
+   }
+}

pipeline/src/data_generation/llama/destructor.cpp

@@ -0,0 +1,31 @@
+/**
+ * Destructor Module
+ * Ensures proper cleanup of llama.cpp resources (context and model) when the
+ * generator is destroyed, preventing memory leaks and resource exhaustion.
+ */
+
+#include "data_generation/llama_generator.h"
+#include "llama.h"
+
+LlamaGenerator::~LlamaGenerator() {
+   /**
+    * Free the inference context (contains KV cache and computation state)
+    */
+   if (context_ != nullptr) {
+      llama_free(context_);
+      context_ = nullptr;
+   }
+
+   /**
+    * Free the loaded model (contains weights and vocabulary)
+    */
+   if (model_ != nullptr) {
+      llama_model_free(model_);
+      model_ = nullptr;
+   }
+
+   /**
+    * Clean up the backend (GPU/CPU acceleration resources)
+    */
+   llama_backend_free();
+}

pipeline/src/data_generation/llama/generate_brewery.cpp

@@ -0,0 +1,107 @@
+/**
+ * Brewery Data Generation Module
+ * Uses the LLM to generate realistic brewery names and descriptions for a given
+ * location. Implements retry logic with validation and error correction to
+ * ensure valid JSON output conforming to the expected schema.
+ */
+
+#include <spdlog/spdlog.h>
+
+#include <stdexcept>
+#include <string>
+
+#include "data_generation/llama_generator.h"
+#include "data_generation/llama_generator_helpers.h"
+
+BreweryResult LlamaGenerator::GenerateBrewery(
+    const std::string& city_name, const std::string& country_name,
+    const std::string& region_context) {
+   /**
+    * Preprocess and truncate region context to manageable size
+    */
+   const std::string safe_region_context =
+       PrepareRegionContextPublic(region_context);
+
+   /**
+    * Load brewery system prompt from file
+    * Falls back to minimal inline prompt if file not found
+    * Default path: prompts/brewery_system_prompt_expanded.txt
+    */
+   const std::string system_prompt =
+       LoadBrewerySystemPrompt("prompts/brewery_system_prompt_expanded.txt");
+
+   /**
+    * User prompt: provides geographic context to guide generation towards
+    * culturally appropriate and locally-inspired brewery attributes
+    */
+   std::string prompt =
+       "Write a brewery name and place-specific long description for a craft "
+       "brewery in " +
+       city_name +
+       (country_name.empty() ? std::string("")
+                             : std::string(", ") + country_name) +
+       (safe_region_context.empty()
+            ? std::string(".")
+            : std::string(". Regional context: ") + safe_region_context);
+
+   /**
+    * Store location context for retry prompts (without repeating full context)
+    */
+   const std::string retry_location =
+       "Location: " + city_name +
+       (country_name.empty() ? std::string("")
+                             : std::string(", ") + country_name);
+
+   /**
+    * RETRY LOOP with validation and error correction
+    * Attempts to generate valid brewery data up to 3 times, with feedback-based
+    * refinement
+    */
+   const int max_attempts = 3;
+   std::string raw;
+   std::string last_error;
+
+   // Limit output length to keep it concise and focused
+   constexpr int max_tokens = 1052;
+   for (int attempt = 0; attempt < max_attempts; ++attempt) {
+      // Generate brewery data from LLM
+      raw = Infer(system_prompt, prompt, max_tokens);
+      spdlog::debug("LlamaGenerator: raw output (attempt {}): {}", attempt + 1,
+                    raw);
+
+      // Validate output: parse JSON and check required fields
+
+      std::string name;
+      std::string description;
+      const std::string validation_error =
+          ValidateBreweryJsonPublic(raw, name, description);
+      if (validation_error.empty()) {
+         // Success: return parsed brewery data
+         return {std::move(name), std::move(description)};
+      }
+
+      // Validation failed: log error and prepare corrective feedback
+
+      last_error = validation_error;
+      spdlog::warn("LlamaGenerator: malformed brewery JSON (attempt {}): {}",
+                   attempt + 1, validation_error);
+
+      // Update prompt with error details to guide LLM toward correct output.
+      // For retries, use a compact prompt format to avoid exceeding token
+      // limits.
+      prompt =
+          "Your previous response was invalid. Error: " + validation_error +
+          "\nReturn ONLY valid JSON with this exact schema: "
+          "{\"name\": \"string\", \"description\": \"string\"}."
+          "\nDo not include markdown, comments, or extra keys."
+          "\n\n" +
+          retry_location;
+   }
+
+   // All retry attempts exhausted: log failure and throw exception
+   spdlog::error(
+       "LlamaGenerator: malformed brewery response after {} attempts: "
+       "{}",
+       max_attempts, last_error.empty() ? raw : last_error);
+   throw std::runtime_error("LlamaGenerator: malformed brewery response");
+}

pipeline/src/data_generation/llama/generate_user.cpp

@@ -0,0 +1,102 @@
+/**
+ * User Profile Generation Module
+ * Uses the LLM to generate realistic user profiles (username and bio) for craft
+ * beer enthusiasts. Implements retry logic to handle parsing failures and
+ * ensures output adheres to strict format constraints (two lines, specific
+ * character limits).
+ */
+
+#include <spdlog/spdlog.h>
+
+#include <algorithm>
+#include <stdexcept>
+#include <string>
+
+#include "data_generation/llama_generator.h"
+#include "data_generation/llama_generator_helpers.h"
+
+UserResult LlamaGenerator::GenerateUser(const std::string& locale) {
+   /**
+    * System prompt: specifies exact output format to minimize parsing errors
+    * Constraints: 2-line output, username format, bio length bounds
+    */
+   const std::string system_prompt =
+       "You generate plausible social media profiles for craft beer "
+       "enthusiasts. "
+       "Respond with exactly two lines: "
+       "the first line is a username (lowercase, no spaces, 8-20 characters), "
+       "the second line is a one-sentence bio (20-40 words). "
+       "The profile should feel consistent with the locale. "
+       "No preamble, no labels.";
+
+   /**
+    * User prompt: locale parameter guides cultural appropriateness of generated
+    * profiles
+    */
+   std::string prompt =
+       "Generate a craft beer enthusiast profile. Locale: " + locale;
+
+   /**
+    * RETRY LOOP with format validation
+    * Attempts up to 3 times to generate valid user profile with correct format
+    */
+   const int max_attempts = 3;
+   std::string raw;
+   for (int attempt = 0; attempt < max_attempts; ++attempt) {
+      /**
+       * Generate user profile (max 128 tokens - should fit 2 lines easily)
+       */
+      raw = Infer(system_prompt, prompt, 128);
+      spdlog::debug("LlamaGenerator (user): raw output (attempt {}): {}",
+                    attempt + 1, raw);
+
+      try {
+         /**
+          * Parse two-line response: first line = username, second line = bio
+          */
+         auto [username, bio] = ParseTwoLineResponsePublic(
+             raw, "LlamaGenerator: malformed user response");
+
+         /**
+          * Remove any whitespace from username (usernames shouldn't have
+          * spaces)
+          */
+         username.erase(
+             std::remove_if(username.begin(), username.end(),
+                            [](unsigned char ch) { return std::isspace(ch); }),
+             username.end());
+
+         /**
+          * Validate both fields are non-empty after processing
+          */
+         if (username.empty() || bio.empty()) {
+            throw std::runtime_error("LlamaGenerator: malformed user response");
+         }
+
+         /**
+          * Truncate bio if exceeds reasonable length for bio field
+          */
+         if (bio.size() > 200) bio = bio.substr(0, 200);
+
+         /**
+          * Success: return parsed user profile
+          */
+         return {username, bio};
+      } catch (const std::exception& e) {
+         /**
+          * Parsing failed: log and continue to next attempt
+          */
+         spdlog::warn(
+             "LlamaGenerator: malformed user response (attempt {}): {}",
+             attempt + 1, e.what());
+      }
+   }
+
+   /**
+    * All retry attempts exhausted: log failure and throw exception
+    */
+   spdlog::error(
+       "LlamaGenerator: malformed user response after {} attempts: {}",
+       max_attempts, raw);
+   throw std::runtime_error("LlamaGenerator: malformed user response");
+}

pipeline/src/data_generation/llama/helpers.cpp

@@ -0,0 +1,441 @@
+/**
+ * Helper Functions Module
+ * Provides utility functions for text processing, parsing, and chat template
+ * formatting. Functions handle whitespace normalization, response parsing, and
+ * conversion of prompts to proper chat format using the model's built-in
+ * template.
+ */
+
+#include <algorithm>
+#include <array>
+#include <boost/json.hpp>
+#include <cctype>
+#include <sstream>
+#include <stdexcept>
+#include <string>
+#include <vector>
+
+#include "data_generation/llama_generator.h"
+#include "llama.h"
+
+namespace {
+
+/**
+ * String trimming: removes leading and trailing whitespace
+ */
+std::string Trim(std::string value) {
+   auto not_space = [](unsigned char ch) { return !std::isspace(ch); };
+
+   value.erase(value.begin(),
+               std::find_if(value.begin(), value.end(), not_space));
+   value.erase(std::find_if(value.rbegin(), value.rend(), not_space).base(),
+               value.end());
+
+   return value;
+}
+
+/**
+ * Normalize whitespace: collapses multiple spaces/tabs/newlines into single
+ * spaces
+ */
+std::string CondenseWhitespace(std::string text) {
+   std::string out;
+   out.reserve(text.size());
+
+   bool in_whitespace = false;
+   for (unsigned char ch : text) {
+      if (std::isspace(ch)) {
+         if (!in_whitespace) {
+            out.push_back(' ');
+            in_whitespace = true;
+         }
+         continue;
+      }
+
+      in_whitespace = false;
+      out.push_back(static_cast<char>(ch));
+   }
+
+   return Trim(std::move(out));
+}
+
+/**
+ * Truncate region context to fit within max length while preserving word
+ * boundaries
+ */
+std::string PrepareRegionContext(std::string_view region_context,
+                                 std::size_t max_chars) {
+   std::string normalized = CondenseWhitespace(std::string(region_context));
+   if (normalized.size() <= max_chars) {
+      return normalized;
+   }
+
+   normalized.resize(max_chars);
+   const std::size_t last_space = normalized.find_last_of(' ');
+   if (last_space != std::string::npos && last_space > max_chars / 2) {
+      normalized.resize(last_space);
+   }
+
+   normalized += "...";
+   return normalized;
+}
+
+/**
+ * Remove common bullet points, numbers, and field labels added by LLM in output
+ */
+std::string StripCommonPrefix(std::string line) {
+   line = Trim(std::move(line));
+
+   if (!line.empty() && (line[0] == '-' || line[0] == '*')) {
+      line = Trim(line.substr(1));
+   } else {
+      std::size_t i = 0;
+      while (i < line.size() &&
+             std::isdigit(static_cast<unsigned char>(line[i]))) {
+         ++i;
+      }
+      if (i > 0 && i < line.size() && (line[i] == '.' || line[i] == ')')) {
+         line = Trim(line.substr(i + 1));
+      }
+   }
+
+   auto strip_label = [&line](const std::string& label) {
+      if (line.size() >= label.size()) {
+         bool matches = true;
+         for (std::size_t i = 0; i < label.size(); ++i) {
+            if (std::tolower(static_cast<unsigned char>(line[i])) !=
+                std::tolower(static_cast<unsigned char>(label[i]))) {
+               matches = false;
+               break;
+            }
+         }
+         if (matches) {
+            line = Trim(line.substr(label.size()));
+         }
+      }
+   };
+
+   strip_label("name:");
+   strip_label("brewery name:");
+   strip_label("description:");
+   strip_label("username:");
+   strip_label("bio:");
+
+   return Trim(std::move(line));
+}
+
+/**
+ * Parse two-line response from LLM: normalize line endings, strip formatting,
+ * filter spurious output, and combine remaining lines if needed
+ */
+std::pair<std::string, std::string> ParseTwoLineResponse(
+    const std::string& raw, const std::string& error_message) {
+   std::string normalized = raw;
+   std::replace(normalized.begin(), normalized.end(), '\r', '\n');
+
+   std::vector<std::string> lines;
+   std::stringstream stream(normalized);
+   std::string line;
+   while (std::getline(stream, line)) {
+      line = StripCommonPrefix(std::move(line));
+      if (!line.empty()) lines.push_back(std::move(line));
+   }
+
+   std::vector<std::string> filtered;
+   for (auto& l : lines) {
+      std::string low = l;
+      std::transform(low.begin(), low.end(), low.begin(), [](unsigned char c) {
+         return static_cast<char>(std::tolower(c));
+      });
+      // Filter known thinking tags like <think>...</think>, but be conservative
+      // to avoid removing legitimate output. Only filter specific known
+      // patterns.
+      if (!l.empty() && l.front() == '<' && low.back() == '>') {
+         // Only filter if it's a known thinking tag: <think>, <reasoning>, etc.
+         if (low.find("think") != std::string::npos ||
+             low.find("reasoning") != std::string::npos ||
+             low.find("reflect") != std::string::npos) {
+            continue;
+         }
+      }
+      if (low.rfind("okay,", 0) == 0 || low.rfind("hmm", 0) == 0) continue;
+      filtered.push_back(std::move(l));
+   }
+
+   if (filtered.size() < 2) throw std::runtime_error(error_message);
+
+   std::string first = Trim(filtered.front());
+   std::string second;
+   for (size_t i = 1; i < filtered.size(); ++i) {
+      if (!second.empty()) second += ' ';
+      second += filtered[i];
+   }
+   second = Trim(std::move(second));
+
+   if (first.empty() || second.empty()) throw std::runtime_error(error_message);
+   return {first, second};
+}
+
+/**
+ * Apply model's chat template to user-only prompt, formatting it for the model
+ */
+std::string ToChatPrompt(const llama_model* model,
+                         const std::string& user_prompt) {
+   const char* tmpl = llama_model_chat_template(model, nullptr);
+   if (tmpl == nullptr) {
+      return user_prompt;
+   }
+
+   const llama_chat_message message{"user", user_prompt.c_str()};
+
+   std::vector<char> buffer(
+       std::max<std::size_t>(1024, user_prompt.size() * 4));
+   int32_t required =
+       llama_chat_apply_template(tmpl, &message, 1, true, buffer.data(),
+                                 static_cast<int32_t>(buffer.size()));
+
+   if (required < 0) {
+      throw std::runtime_error("LlamaGenerator: failed to apply chat template");
+   }
+
+   if (required >= static_cast<int32_t>(buffer.size())) {
+      buffer.resize(static_cast<std::size_t>(required) + 1);
+      required =
+          llama_chat_apply_template(tmpl, &message, 1, true, buffer.data(),
+                                    static_cast<int32_t>(buffer.size()));
+      if (required < 0) {
+         throw std::runtime_error(
+             "LlamaGenerator: failed to apply chat template");
+      }
+   }
+
+   return std::string(buffer.data(), static_cast<std::size_t>(required));
+}
+
+/**
+ * Apply model's chat template to system+user prompt pair, formatting for the
+ * model
+ */
+std::string ToChatPrompt(const llama_model* model,
+                         const std::string& system_prompt,
+                         const std::string& user_prompt) {
+   const char* tmpl = llama_model_chat_template(model, nullptr);
+   if (tmpl == nullptr) {
+      return system_prompt + "\n\n" + user_prompt;
+   }
+
+   const llama_chat_message messages[2] = {{"system", system_prompt.c_str()},
+                                           {"user", user_prompt.c_str()}};
+
+   std::vector<char> buffer(std::max<std::size_t>(
+       1024, (system_prompt.size() + user_prompt.size()) * 4));
+   int32_t required =
+       llama_chat_apply_template(tmpl, messages, 2, true, buffer.data(),
+                                 static_cast<int32_t>(buffer.size()));
+
+   if (required < 0) {
+      throw std::runtime_error("LlamaGenerator: failed to apply chat template");
+   }
+
+   if (required >= static_cast<int32_t>(buffer.size())) {
+      buffer.resize(static_cast<std::size_t>(required) + 1);
+      required =
+          llama_chat_apply_template(tmpl, messages, 2, true, buffer.data(),
+                                    static_cast<int32_t>(buffer.size()));
+      if (required < 0) {
+         throw std::runtime_error(
+             "LlamaGenerator: failed to apply chat template");
+      }
+   }
+
+   return std::string(buffer.data(), static_cast<std::size_t>(required));
+}
+
+void AppendTokenPiece(const llama_vocab* vocab, llama_token token,
+                      std::string& output) {
+   std::array<char, 256> buffer{};
+   int32_t bytes =
+       llama_token_to_piece(vocab, token, buffer.data(),
+                            static_cast<int32_t>(buffer.size()), 0, true);
+
+   if (bytes < 0) {
+      std::vector<char> dynamic_buffer(static_cast<std::size_t>(-bytes));
+      bytes = llama_token_to_piece(vocab, token, dynamic_buffer.data(),
+                                   static_cast<int32_t>(dynamic_buffer.size()),
+                                   0, true);
+      if (bytes < 0) {
+         throw std::runtime_error(
+             "LlamaGenerator: failed to decode sampled token piece");
+      }
+
+      output.append(dynamic_buffer.data(), static_cast<std::size_t>(bytes));
+      return;
+   }
+
+   output.append(buffer.data(), static_cast<std::size_t>(bytes));
+}
+
+bool ExtractFirstJsonObject(const std::string& text, std::string& json_out) {
+   std::size_t start = std::string::npos;
+   int depth = 0;
+   bool in_string = false;
+   bool escaped = false;
+
+   for (std::size_t i = 0; i < text.size(); ++i) {
+      const char ch = text[i];
+
+      if (in_string) {
+         if (escaped) {
+            escaped = false;
+         } else if (ch == '\\') {
+            escaped = true;
+         } else if (ch == '"') {
+            in_string = false;
+         }
+         continue;
+      }
+
+      if (ch == '"') {
+         in_string = true;
+         continue;
+      }
+
+      if (ch == '{') {
+         if (depth == 0) {
+            start = i;
+         }
+         ++depth;
+         continue;
+      }
+
+      if (ch == '}') {
+         if (depth == 0) {
+            continue;
+         }
+         --depth;
+         if (depth == 0 && start != std::string::npos) {
+            json_out = text.substr(start, i - start + 1);
+            return true;
+         }
+      }
+   }
+
+   return false;
+}
+
+std::string ValidateBreweryJson(const std::string& raw, std::string& name_out,
+                                std::string& description_out) {
+   auto validate_object = [&](const boost::json::value& jv,
+                              std::string& error_out) -> bool {
+      if (!jv.is_object()) {
+         error_out = "JSON root must be an object";
+         return false;
+      }
+
+      const auto& obj = jv.get_object();
+      if (!obj.contains("name") || !obj.at("name").is_string()) {
+         error_out = "JSON field 'name' is missing or not a string";
+         return false;
+      }
+
+      if (!obj.contains("description") || !obj.at("description").is_string()) {
+         error_out = "JSON field 'description' is missing or not a string";
+         return false;
+      }
+
+      name_out = Trim(std::string(obj.at("name").as_string().c_str()));
+      description_out =
+          Trim(std::string(obj.at("description").as_string().c_str()));
+
+      if (name_out.empty()) {
+         error_out = "JSON field 'name' must not be empty";
+         return false;
+      }
+
+      if (description_out.empty()) {
+         error_out = "JSON field 'description' must not be empty";
+         return false;
+      }
+
+      std::string name_lower = name_out;
+      std::string description_lower = description_out;
+      std::transform(
+          name_lower.begin(), name_lower.end(), name_lower.begin(),
+          [](unsigned char c) { return static_cast<char>(std::tolower(c)); });
+      std::transform(description_lower.begin(), description_lower.end(),
+                     description_lower.begin(), [](unsigned char c) {
+                        return static_cast<char>(std::tolower(c));
+                     });
+
+      if (name_lower == "string" || description_lower == "string") {
+         error_out = "JSON appears to be a schema placeholder, not content";
+         return false;
+      }
+
+      error_out.clear();
+      return true;
+   };
+
+   boost::system::error_code ec;
+   boost::json::value jv = boost::json::parse(raw, ec);
+   std::string validation_error;
+   if (ec) {
+      std::string extracted;
+      if (!ExtractFirstJsonObject(raw, extracted)) {
+         return "JSON parse error: " + ec.message();
+      }
+
+      ec.clear();
+      jv = boost::json::parse(extracted, ec);
+      if (ec) {
+         return "JSON parse error: " + ec.message();
+      }
+
+      if (!validate_object(jv, validation_error)) {
+         return validation_error;
+      }
+
+      return {};
+   }
+
+   if (!validate_object(jv, validation_error)) {
+      return validation_error;
+   }
+
+   return {};
+}
+
+}  // namespace
+
+// Forward declarations for helper functions exposed to other translation units
+std::string PrepareRegionContextPublic(std::string_view region_context,
+                                       std::size_t max_chars) {
+   return PrepareRegionContext(region_context, max_chars);
+}
+
+std::pair<std::string, std::string> ParseTwoLineResponsePublic(
+    const std::string& raw, const std::string& error_message) {
+   return ParseTwoLineResponse(raw, error_message);
+}
+
+std::string ToChatPromptPublic(const llama_model* model,
+                               const std::string& user_prompt) {
+   return ToChatPrompt(model, user_prompt);
+}
+
+std::string ToChatPromptPublic(const llama_model* model,
+                               const std::string& system_prompt,
+                               const std::string& user_prompt) {
+   return ToChatPrompt(model, system_prompt, user_prompt);
+}
+
+void AppendTokenPiecePublic(const llama_vocab* vocab, llama_token token,
+                            std::string& output) {
+   AppendTokenPiece(vocab, token, output);
+}
+
+std::string ValidateBreweryJsonPublic(const std::string& raw,
+                                      std::string& name_out,
+                                      std::string& description_out) {
+   return ValidateBreweryJson(raw, name_out, description_out);
+}

pipeline/src/data_generation/llama/infer.cpp

@@ -0,0 +1,196 @@
+/**
+ * Text Generation / Inference Module
+ * Core module that performs LLM inference: converts text prompts into tokens,
+ * runs the neural network forward pass, samples the next token, and converts
+ * output tokens back to text. Supports both simple and system+user prompts.
+ */
+
+#include <spdlog/spdlog.h>
+
+#include <algorithm>
+#include <memory>
+#include <stdexcept>
+#include <string>
+#include <vector>
+
+#include "data_generation/llama_generator.h"
+#include "data_generation/llama_generator_helpers.h"
+#include "llama.h"
+
+std::string LlamaGenerator::Infer(const std::string& prompt, int max_tokens) {
+   return InferFormatted(ToChatPromptPublic(model_, prompt), max_tokens);
+}
+
+std::string LlamaGenerator::Infer(const std::string& system_prompt,
+                                  const std::string& prompt, int max_tokens) {
+   return InferFormatted(ToChatPromptPublic(model_, system_prompt, prompt),
+                         max_tokens);
+}
+
+std::string LlamaGenerator::InferFormatted(const std::string& formatted_prompt,
+                                           int max_tokens) {
+   /**
+    * Validate that model and context are loaded
+    */
+   if (model_ == nullptr || context_ == nullptr)
+      throw std::runtime_error("LlamaGenerator: model not loaded");
+
+   /**
+    * Get vocabulary for tokenization and token-to-text conversion
+    */
+   const llama_vocab* vocab = llama_model_get_vocab(model_);
+   if (vocab == nullptr)
+      throw std::runtime_error("LlamaGenerator: vocab unavailable");
+
+   /**
+    * Clear KV cache to ensure clean inference state (no residual context)
+    */
+   llama_memory_clear(llama_get_memory(context_), true);
+
+   /**
+    * TOKENIZATION PHASE
+    * Convert text prompt into token IDs (integers) that the model understands
+    */
+   std::vector<llama_token> prompt_tokens(formatted_prompt.size() + 8);
+   int32_t token_count = llama_tokenize(
+       vocab, formatted_prompt.c_str(),
+       static_cast<int32_t>(formatted_prompt.size()), prompt_tokens.data(),
+       static_cast<int32_t>(prompt_tokens.size()), true, true);
+
+   /**
+    * If buffer too small, negative return indicates required size
+    */
+   if (token_count < 0) {
+      prompt_tokens.resize(static_cast<std::size_t>(-token_count));
+      token_count = llama_tokenize(
+          vocab, formatted_prompt.c_str(),
+          static_cast<int32_t>(formatted_prompt.size()), prompt_tokens.data(),
+          static_cast<int32_t>(prompt_tokens.size()), true, true);
+   }
+
+   if (token_count < 0)
+      throw std::runtime_error("LlamaGenerator: prompt tokenization failed");
+
+   /**
+    * CONTEXT SIZE VALIDATION
+    * Validate and compute effective token budgets based on context window
+    * constraints
+    */
+   const int32_t n_ctx = static_cast<int32_t>(llama_n_ctx(context_));
+   const int32_t n_batch = static_cast<int32_t>(llama_n_batch(context_));
+   if (n_ctx <= 1 || n_batch <= 0)
+      throw std::runtime_error("LlamaGenerator: invalid context or batch size");
+
+   /**
+    * Clamp generation limit to available context window, reserve space for
+    * output
+    */
+   const int32_t effective_max_tokens =
+       std::max(1, std::min(max_tokens, n_ctx - 1));
+   /**
+    * Prompt can use remaining context after reserving space for generation
+    */
+   int32_t prompt_budget = std::min(n_batch, n_ctx - effective_max_tokens);
+   prompt_budget = std::max<int32_t>(1, prompt_budget);
+
+   /**
+    * Truncate prompt if necessary to fit within constraints
+    */
+   prompt_tokens.resize(static_cast<std::size_t>(token_count));
+   if (token_count > prompt_budget) {
+      spdlog::warn(
+          "LlamaGenerator: prompt too long ({} tokens), truncating to {} "
+          "tokens to fit n_batch/n_ctx limits",
+          token_count, prompt_budget);
+      prompt_tokens.resize(static_cast<std::size_t>(prompt_budget));
+      token_count = prompt_budget;
+   }
+
+   /**
+    * PROMPT PROCESSING PHASE
+    * Create a batch containing all prompt tokens and feed through the model
+    * This computes internal representations and fills the KV cache
+    */
+   const llama_batch prompt_batch = llama_batch_get_one(
+       prompt_tokens.data(), static_cast<int32_t>(prompt_tokens.size()));
+   if (llama_decode(context_, prompt_batch) != 0)
+      throw std::runtime_error("LlamaGenerator: prompt decode failed");
+
+   /**
+    * SAMPLER CONFIGURATION PHASE
+    * Set up the probabilistic token selection pipeline (sampler chain)
+    * Samplers are applied in sequence: temperature -> top-p -> distribution
+    */
+   llama_sampler_chain_params sampler_params =
+       llama_sampler_chain_default_params();
+   using SamplerPtr =
+       std::unique_ptr<llama_sampler, decltype(&llama_sampler_free)>;
+   SamplerPtr sampler(llama_sampler_chain_init(sampler_params),
+                      &llama_sampler_free);
+   if (!sampler)
+      throw std::runtime_error("LlamaGenerator: failed to initialize sampler");
+
+   /**
+    * Temperature: scales logits before softmax (controls randomness)
+    */
+   llama_sampler_chain_add(sampler.get(),
+                           llama_sampler_init_temp(sampling_temperature_));
+   /**
+    * Top-P: nucleus sampling - filters to most likely tokens summing to top_p
+    * probability
+    */
+   llama_sampler_chain_add(sampler.get(),
+                           llama_sampler_init_top_p(sampling_top_p_, 1));
+   /**
+    * Distribution sampler: selects actual token using configured seed for
+    * reproducibility
+    */
+   llama_sampler_chain_add(sampler.get(),
+                           llama_sampler_init_dist(sampling_seed_));
+
+   /**
+    * TOKEN GENERATION LOOP
+    * Iteratively generate tokens one at a time until max_tokens or
+    * end-of-sequence
+    */
+   std::vector<llama_token> generated_tokens;
+   generated_tokens.reserve(static_cast<std::size_t>(effective_max_tokens));
+
+   for (int i = 0; i < effective_max_tokens; ++i) {
+      /**
+       * Sample next token using configured sampler chain and model logits
+       * Index -1 means use the last output position from previous batch
+       */
+      const llama_token next =
+          llama_sampler_sample(sampler.get(), context_, -1);
+      /**
+       * Stop if model predicts end-of-generation token (EOS/EOT)
+       */
+      if (llama_vocab_is_eog(vocab, next)) break;
+      generated_tokens.push_back(next);
+      /**
+       * Feed the sampled token back into model for next iteration
+       * (autoregressive)
+       */
+      llama_token token = next;
+      const llama_batch one_token_batch = llama_batch_get_one(&token, 1);
+      if (llama_decode(context_, one_token_batch) != 0)
+         throw std::runtime_error(
+             "LlamaGenerator: decode failed during generation");
+   }
+
+   /**
+    * DETOKENIZATION PHASE
+    * Convert generated token IDs back to text using vocabulary
+    */
+   std::string output;
+   for (const llama_token token : generated_tokens)
+      AppendTokenPiecePublic(vocab, token, output);
+
+   /**
+    * Advance seed for next generation to improve output diversity
+    */
+   sampling_seed_ = (sampling_seed_ == 0xFFFFFFFFu) ? 0 : sampling_seed_ + 1;
+
+   return output;
+}

pipeline/src/data_generation/llama/load.cpp

@@ -0,0 +1,56 @@
+/**
+ * Model Loading Module
+ * This module handles loading a pre-trained LLM model from disk and
+ * initializing the llama.cpp context for inference. It performs one-time setup
+ * required before any inference operations can be performed.
+ */
+
+#include <spdlog/spdlog.h>
+
+#include <stdexcept>
+#include <string>
+
+#include "data_generation/llama_generator.h"
+#include "llama.h"
+
+void LlamaGenerator::Load(const std::string& model_path) {
+   /**
+    * Validate input and clean up any previously loaded model/context
+    */
+   if (model_path.empty())
+      throw std::runtime_error("LlamaGenerator: model path must not be empty");
+
+   if (context_ != nullptr) {
+      llama_free(context_);
+      context_ = nullptr;
+   }
+   if (model_ != nullptr) {
+      llama_model_free(model_);
+      model_ = nullptr;
+   }
+
+   /**
+    * Initialize the llama backend (one-time setup for GPU/CPU acceleration)
+    */
+   llama_backend_init();
+
+   llama_model_params model_params = llama_model_default_params();
+   model_ = llama_model_load_from_file(model_path.c_str(), model_params);
+   if (model_ == nullptr) {
+      throw std::runtime_error(
+          "LlamaGenerator: failed to load model from path: " + model_path);
+   }
+
+   llama_context_params context_params = llama_context_default_params();
+   context_params.n_ctx = n_ctx_;
+   context_params.n_batch = n_ctx_;  // Set batch size equal to context window
+
+   context_ = llama_init_from_model(model_, context_params);
+   if (context_ == nullptr) {
+      llama_model_free(model_);
+      model_ = nullptr;
+      throw std::runtime_error("LlamaGenerator: failed to create context");
+   }
+
+   spdlog::info("[LlamaGenerator] Loaded model: {}", model_path);
+}

pipeline/src/data_generation/llama/load_brewery_prompt.cpp

@@ -0,0 +1,74 @@
+#include <fstream>
+#include <filesystem>
+#include <spdlog/spdlog.h>
+
+#include "data_generation/llama_generator.h"
+
+namespace fs = std::filesystem;
+
+std::string LlamaGenerator::LoadBrewerySystemPrompt(
+    const std::string& prompt_file_path) {
+   // Return cached version if already loaded
+   if (!brewery_system_prompt_.empty()) {
+      return brewery_system_prompt_;
+   }
+
+   // Try multiple path locations
+   std::vector<std::string> paths_to_try = {
+       prompt_file_path,                          // As provided
+       "../" + prompt_file_path,                  // One level up
+       "../../" + prompt_file_path,               // Two levels up
+   };
+
+   for (const auto& path : paths_to_try) {
+      std::ifstream prompt_file(path);
+      if (prompt_file.is_open()) {
+         std::string prompt((std::istreambuf_iterator<char>(prompt_file)),
+                            std::istreambuf_iterator<char>());
+         prompt_file.close();
+
+         if (!prompt.empty()) {
+            spdlog::info(
+                "LlamaGenerator: Loaded brewery system prompt from '{}' ({} chars)",
+                path, prompt.length());
+            brewery_system_prompt_ = prompt;
+            return brewery_system_prompt_;
+         }
+      }
+   }
+
+   spdlog::warn(
+       "LlamaGenerator: Could not open brewery system prompt file at any of the "
+       "expected locations. Using fallback inline prompt.");
+   return GetFallbackBreweryPrompt();
+}
+
+// Fallback: minimal inline prompt if file fails to load
+std::string LlamaGenerator::GetFallbackBreweryPrompt() {
+   return "You are an experienced brewmaster and owner of a local craft brewery. "
+          "Create a distinctive, authentic name and detailed description that "
+          "genuinely reflects your specific location, brewing philosophy, local "
+          "culture, and community connection. The brewery must feel real and "
+          "grounded—not generic or interchangeable.\n\n"
+          "AVOID REPETITIVE PHRASES - Never use:\n"
+          "Love letter to, tribute to, rolling hills, picturesque, every sip "
+          "tells a story, Come for X stay for Y, rich history, passion, woven "
+          "into, ancient roots, timeless, where tradition meets innovation\n\n"
+          "OPENING APPROACHES - Choose ONE:\n"
+          "1. Start with specific beer style and its regional origins\n"
+          "2. Begin with specific brewing challenge (water, altitude, climate)\n"
+          "3. Open with founding story or personal motivation\n"
+          "4. Lead with specific local ingredient or resource\n"
+          "5. Start with unexpected angle or contradiction\n"
+          "6. Open with local event, tradition, or cultural moment\n"
+          "7. Begin with tangible architectural or geographic detail\n\n"
+          "BE SPECIFIC - Include:\n"
+          "- At least ONE concrete proper noun (landmark, river, neighborhood)\n"
+          "- Specific beer styles relevant to the REGION'S culture\n"
+          "- Concrete brewing challenges or advantages\n"
+          "- Sensory details SPECIFIC to place—not generic adjectives\n\n"
+          "LENGTH: 150-250 words. TONE: Can be soulful, irreverent, "
+          "matter-of-fact, unpretentious, or minimalist.\n\n"
+          "Output ONLY a raw JSON object with keys name and description. "
+          "No markdown, backticks, preamble, or trailing text.";
+}

pipeline/src/data_generation/llama/set_sampling_options.cpp

@@ -0,0 +1,65 @@
+/**
+ * Sampling Configuration Module
+ * Configures the hyperparameters that control probabilistic token selection
+ * during text generation. These settings affect the randomness, diversity, and
+ * quality of generated output.
+ */
+
+#include <stdexcept>
+
+#include "data_generation/llama_generator.h"
+#include "llama.h"
+
+void LlamaGenerator::SetSamplingOptions(float temperature, float top_p,
+                                        int seed) {
+   /**
+    * Validate temperature: controls randomness in output distribution
+    * 0.0 = deterministic (always pick highest probability token)
+    * Higher values = more random/diverse output
+    */
+   if (temperature < 0.0f) {
+      throw std::runtime_error(
+          "LlamaGenerator: sampling temperature must be >= 0");
+   }
+
+   /**
+    * Validate top-p (nucleus sampling): only sample from top cumulative
+    * probability e.g., top-p=0.9 means sample from tokens that make up 90% of
+    * probability mass
+    */
+   if (!(top_p > 0.0f && top_p <= 1.0f)) {
+      throw std::runtime_error(
+          "LlamaGenerator: sampling top-p must be in (0, 1]");
+   }
+
+   /**
+    * Validate seed: for reproducible results (-1 uses random seed)
+    */
+   if (seed < -1) {
+      throw std::runtime_error(
+          "LlamaGenerator: seed must be >= 0, or -1 for random");
+   }
+
+   /**
+    * Store sampling parameters for use during token generation
+    */
+   sampling_temperature_ = temperature;
+   sampling_top_p_ = top_p;
+   sampling_seed_ = (seed < 0) ? static_cast<uint32_t>(LLAMA_DEFAULT_SEED)
+                               : static_cast<uint32_t>(seed);
+}
+
+void LlamaGenerator::SetContextSize(uint32_t n_ctx) {
+   /**
+    * Validate context size: must be positive and reasonable for the model
+    */
+   if (n_ctx == 0 || n_ctx > 32768) {
+      throw std::runtime_error(
+          "LlamaGenerator: context size must be in range [1, 32768]");
+   }
+
+   /**
+    * Store context size for use during model loading
+    */
+   n_ctx_ = n_ctx;
+}

pipeline/src/data_generation/mock/data.cpp

@@ -0,0 +1,65 @@
+#include <string>
+#include <vector>
+
+#include "data_generation/mock_generator.h"
+
+const std::vector<std::string> MockGenerator::kBreweryAdjectives = {
+    "Craft",      "Heritage", "Local",  "Artisan",  "Pioneer",    "Golden",
+    "Modern",     "Classic",  "Summit", "Northern", "Riverstone", "Barrel",
+    "Hinterland", "Harbor",   "Wild",   "Granite",  "Copper",     "Maple"};
+
+const std::vector<std::string> MockGenerator::kBreweryNouns = {
+    "Brewing Co.", "Brewery",    "Bier Haus", "Taproom",      "Works",
+    "House",       "Fermentery", "Ale Co.",   "Cellars",      "Collective",
+    "Project",     "Foundry",    "Malthouse", "Public House", "Co-op",
+    "Lab",         "Beer Hall",  "Guild"};
+
+const std::vector<std::string> MockGenerator::kBreweryDescriptions = {
+    "Handcrafted pale ales and seasonal IPAs with local ingredients.",
+    "Traditional lagers and experimental sours in small batches.",
+    "Award-winning stouts and wildly hoppy blonde ales.",
+    "Craft brewery specializing in Belgian-style triples and dark porters.",
+    "Modern brewery blending tradition with bold experimental flavors.",
+    "Neighborhood-focused taproom pouring crisp pilsners and citrusy pale "
+    "ales.",
+    "Small-batch brewery known for barrel-aged releases and smoky lagers.",
+    "Independent brewhouse pairing farmhouse ales with rotating food pop-ups.",
+    "Community brewpub making balanced bitters, saisons, and hazy IPAs.",
+    "Experimental nanobrewery exploring local yeast and regional grains.",
+    "Family-run brewery producing smooth amber ales and robust porters.",
+    "Urban brewery crafting clean lagers and bright, fruit-forward sours.",
+    "Riverfront brewhouse featuring oak-matured ales and seasonal blends.",
+    "Modern taproom focused on sessionable lagers and classic pub styles.",
+    "Brewery rooted in tradition with a lineup of malty reds and crisp lagers.",
+    "Creative brewery offering rotating collaborations and limited draft-only "
+    "pours.",
+    "Locally inspired brewery serving approachable ales with bold hop "
+    "character.",
+    "Destination taproom known for balanced IPAs and cocoa-rich stouts."};
+
+const std::vector<std::string> MockGenerator::kUsernames = {
+    "hopseeker",     "malttrail",   "yeastwhisper",  "lagerlane",
+    "barrelbound",   "foamfinder",  "taphunter",     "graingeist",
+    "brewscout",     "aleatlas",    "caskcompass",   "hopsandmaps",
+    "mashpilot",     "pintnomad",   "fermentfriend", "stoutsignal",
+    "sessionwander", "kettlekeeper"};
+
+const std::vector<std::string> MockGenerator::kBios = {
+    "Always chasing balanced IPAs and crisp lagers across local taprooms.",
+    "Weekend brewery explorer with a soft spot for dark, roasty stouts.",
+    "Documenting tiny brewpubs, fresh pours, and unforgettable beer gardens.",
+    "Fan of farmhouse ales, food pairings, and long tasting flights.",
+    "Collecting favorite pilsners one city at a time.",
+    "Hops-first drinker who still saves room for classic malt-forward styles.",
+    "Finding hidden tap lists and sharing the best seasonal releases.",
+    "Brewery road-tripper focused on local ingredients and clean fermentation.",
+    "Always comparing house lagers and ranking patio pint vibes.",
+    "Curious about yeast strains, barrel programs, and cellar experiments.",
+    "Believes every neighborhood deserves a great community taproom.",
+    "Looking for session beers that taste great from first sip to last.",
+    "Belgian ale enthusiast who never skips a new saison.",
+    "Hazy IPA critic with deep respect for a perfectly clear pilsner.",
+    "Visits breweries for the stories, stays for the flagship pours.",
+    "Craft beer fan mapping tasting notes and favorite brew routes.",
+    "Always ready to trade recommendations for underrated local breweries.",
+    "Keeping a running list of must-try collab releases and tap takeovers."};

pipeline/src/data_generation/mock/deterministic_hash.cpp

@@ -0,0 +1,12 @@
+#include <string>
+
+#include "data_generation/mock_generator.h"
+
+std::size_t MockGenerator::DeterministicHash(const std::string& a,
+                                             const std::string& b) {
+   std::size_t seed = std::hash<std::string>{}(a);
+   const std::size_t mixed = std::hash<std::string>{}(b);
+   seed ^= mixed + 0x9e3779b97f4a7c15ULL + (seed << 6) + (seed >> 2);
+   seed = (seed << 13) | (seed >> ((sizeof(std::size_t) * 8) - 13));
+   return seed;
+}

pipeline/src/data_generation/mock/generate_brewery.cpp

@@ -0,0 +1,21 @@
+#include <functional>
+#include <string>
+
+#include "data_generation/mock_generator.h"
+
+BreweryResult MockGenerator::GenerateBrewery(
+    const std::string& city_name, const std::string& country_name,
+    const std::string& region_context) {
+   const std::string location_key =
+       country_name.empty() ? city_name : city_name + "," + country_name;
+   const std::size_t hash =
+       region_context.empty() ? std::hash<std::string>{}(location_key)
+                              : DeterministicHash(location_key, region_context);
+
+   BreweryResult result;
+   result.name = kBreweryAdjectives[hash % kBreweryAdjectives.size()] + " " +
+                 kBreweryNouns[(hash / 7) % kBreweryNouns.size()];
+   result.description =
+       kBreweryDescriptions[(hash / 13) % kBreweryDescriptions.size()];
+   return result;
+}

pipeline/src/data_generation/mock/generate_user.cpp

@@ -0,0 +1,13 @@
+#include <functional>
+#include <string>
+
+#include "data_generation/mock_generator.h"
+
+UserResult MockGenerator::GenerateUser(const std::string& locale) {
+   const std::size_t hash = std::hash<std::string>{}(locale);
+
+   UserResult result;
+   result.username = kUsernames[hash % kUsernames.size()];
+   result.bio = kBios[(hash / 11) % kBios.size()];
+   return result;
+}

pipeline/src/data_generation/mock/load.cpp

@@ -0,0 +1,9 @@
+#include <spdlog/spdlog.h>
+
+#include <string>
+
+#include "data_generation/mock_generator.h"
+
+void MockGenerator::Load(const std::string& /*modelPath*/) {
+   spdlog::info("[MockGenerator] No model needed");
+}

pipeline/src/json_handling/json_loader.cpp

@@ -0,0 +1,83 @@
+#include "json_handling/json_loader.h"
+
+#include <spdlog/spdlog.h>
+
+#include <boost/json.hpp>
+
+#include <fstream>
+#include <sstream>
+#include <stdexcept>
+
+namespace {
+
+auto ReadRequiredString(const boost::json::object& object,
+                        const char* key) -> std::string {
+   const boost::json::value* value = object.if_contains(key);
+   if (value == nullptr || !value->is_string()) {
+      throw std::runtime_error(std::string("Missing or invalid string field: ") +
+                               key);
+   }
+   return std::string(value->as_string().c_str());
+}
+
+auto ReadRequiredNumber(const boost::json::object& object, const char* key)
+    -> double {
+   const boost::json::value* value = object.if_contains(key);
+   if (value == nullptr || !value->is_number()) {
+      throw std::runtime_error(std::string("Missing or invalid numeric field: ") +
+                               key);
+   }
+   return value->to_number<double>();
+}
+
+}  // namespace
+
+auto JsonLoader::LoadLocations(const std::string& filepath)
+   -> std::vector<Location> {
+   std::ifstream input(filepath);
+   if (!input.is_open()) {
+      throw std::runtime_error("Failed to open locations file: " + filepath);
+   }
+
+   std::stringstream buffer;
+   buffer << input.rdbuf();
+   const std::string content = buffer.str();
+
+   boost::json::error_code error;
+   boost::json::value root = boost::json::parse(content, error);
+   if (error) {
+      throw std::runtime_error("Failed to parse locations JSON: " +
+                               error.message());
+   }
+
+   if (!root.is_array()) {
+      throw std::runtime_error(
+          "Invalid locations JSON: root element must be an array");
+   }
+
+   std::vector<Location> locations;
+   const auto& items = root.as_array();
+   locations.reserve(items.size());
+
+   for (const auto& item : items) {
+      if (!item.is_object()) {
+         throw std::runtime_error(
+             "Invalid locations JSON: each entry must be an object");
+      }
+
+      const auto& object = item.as_object();
+      locations.push_back(Location{
+          .city = ReadRequiredString(object, "city"),
+          .state_province = ReadRequiredString(object, "state_province"),
+          .iso3166_2 = ReadRequiredString(object, "iso3166_2"),
+          .country = ReadRequiredString(object, "country"),
+          .iso3166_1 = ReadRequiredString(object, "iso3166_1"),
+          .latitude = ReadRequiredNumber(object, "latitude"),
+          .longitude = ReadRequiredNumber(object, "longitude"),
+      });
+   }
+
+   spdlog::info("[JsonLoader] Loaded {} locations from {}", locations.size(),
+                filepath);
+   return locations;
+}

pipeline/src/main.cpp

@@ -0,0 +1,141 @@
+#include <spdlog/spdlog.h>
+
+#include <boost/program_options.hpp>
+#include <iostream>
+#include <memory>
+
+#include "biergarten_data_generator.h"
+#include "web_client/curl_web_client.h"
+
+namespace po = boost::program_options;
+
+/**
+ * @brief Parse command-line arguments into ApplicationOptions.
+ *
+ * @param argc Command-line argument count.
+ * @param argv Command-line arguments.
+ * @param options Output ApplicationOptions struct.
+ * @return true if parsing succeeded and should proceed, false otherwise.
+ */
+bool ParseArguments(int argc, char** argv, ApplicationOptions& options) {
+   // If no arguments provided, display usage and exit
+   if (argc == 1) {
+      std::cout << "Biergarten Pipeline - Geographic Data Pipeline with "
+                   "Brewery Generation\n\n";
+      std::cout << "Usage: biergarten-pipeline [options]\n\n";
+      std::cout << "Options:\n";
+      std::cout << "  --mocked             Use mocked generator for "
+                   "brewery/user data\n";
+      std::cout << "  --model, -m PATH     Path to LLM model file (gguf) for "
+                   "generation\n";
+      std::cout << "  --cache-dir, -c DIR  Directory for cached JSON (default: "
+                   "/tmp)\n";
+      std::cout << "  --temperature TEMP   LLM sampling temperature 0.0-1.0 "
+                   "(default: 0.8)\n";
+      std::cout << "  --top-p VALUE        Nucleus sampling parameter 0.0-1.0 "
+                   "(default: 0.92)\n";
+      std::cout << "  --n-ctx SIZE         Context window size in tokens "
+                   "(default: 4096)\n";
+      std::cout << "  --seed SEED          Random seed: -1 for random "
+                   "(default: -1)\n";
+      std::cout << "  --help, -h           Show this help message\n\n";
+      std::cout << "Note: --mocked and --model are mutually exclusive. Exactly "
+                   "one must be provided.\n";
+      std::cout << "Data source is always pinned to commit c5eb7772 (stable "
+                   "2026-03-28).\n";
+      return false;
+   }
+
+   po::options_description desc("Pipeline Options");
+   desc.add_options()("help,h", "Produce help message")(
+       "mocked", po::bool_switch(),
+       "Use mocked generator for brewery/user data")(
+       "model,m", po::value<std::string>()->default_value(""),
+       "Path to LLM model (gguf)")(
+       "cache-dir,c", po::value<std::string>()->default_value("/tmp"),
+       "Directory for cached JSON")(
+       "temperature", po::value<float>()->default_value(0.8f),
+       "Sampling temperature (higher = more random)")(
+       "top-p", po::value<float>()->default_value(0.92f),
+       "Nucleus sampling top-p in (0,1] (higher = more random)")(
+       "n-ctx", po::value<uint32_t>()->default_value(8192),
+       "Context window size in tokens (1-32768)")(
+       "seed", po::value<int>()->default_value(-1),
+       "Sampler seed: -1 for random, otherwise non-negative integer");
+
+   po::variables_map vm;
+   po::store(po::parse_command_line(argc, argv, desc), vm);
+   po::notify(vm);
+
+   if (vm.count("help")) {
+      std::cout << desc << "\n";
+      return false;
+   }
+
+   // Check for mutually exclusive --mocked and --model flags
+   bool use_mocked = vm["mocked"].as<bool>();
+   std::string model_path = vm["model"].as<std::string>();
+
+   if (use_mocked && !model_path.empty()) {
+      spdlog::error("ERROR: --mocked and --model are mutually exclusive");
+      return false;
+   }
+
+   if (!use_mocked && model_path.empty()) {
+      spdlog::error("ERROR: Either --mocked or --model must be specified");
+      return false;
+   }
+
+   // Warn if sampling parameters are provided with --mocked
+   if (use_mocked) {
+      bool hasTemperature = vm["temperature"].defaulted() == false;
+      bool hasTopP = vm["top-p"].defaulted() == false;
+      bool hasSeed = vm["seed"].defaulted() == false;
+
+      if (hasTemperature || hasTopP || hasSeed) {
+         spdlog::warn(
+             "WARNING: Sampling parameters (--temperature, --top-p, --seed) "
+             "are ignored when using --mocked");
+      }
+   }
+
+   options.use_mocked = use_mocked;
+   options.model_path = model_path;
+   options.cache_dir = vm["cache-dir"].as<std::string>();
+   options.temperature = vm["temperature"].as<float>();
+   options.top_p = vm["top-p"].as<float>();
+   options.n_ctx = vm["n-ctx"].as<uint32_t>();
+   options.seed = vm["seed"].as<int>();
+   // commit is always pinned to c5eb7772
+
+   return true;
+}
+
+int main(int argc, char* argv[]) {
+   try {
+      const CurlGlobalState curl_state;
+
+      ApplicationOptions options;
+      if (!ParseArguments(argc, argv, options)) {
+         return 0;
+      }
+
+      auto webClient = std::make_shared<CURLWebClient>();
+
+      BiergartenDataGenerator generator(options, webClient);
+      return generator.Run();
+
+   } catch (const std::exception& e) {
+      const std::string message = e.what() ? e.what() : "";
+
+      if (message.find("LlamaGenerator: malformed brewery response") !=
+          std::string::npos) {
+         spdlog::warn("WARNING: Non-fatal LLM failure after retries: {}",
+                      message);
+         return 0;
+      }
+
+      spdlog::error("ERROR: Application failed: {}", e.what());
+      return 1;
+   }
+}

pipeline/src/web_client/curl_web_client.cpp

@@ -0,0 +1,141 @@
+#include "web_client/curl_web_client.h"
+
+#include <curl/curl.h>
+
+#include <cstdio>
+#include <fstream>
+#include <memory>
+#include <sstream>
+#include <stdexcept>
+
+CurlGlobalState::CurlGlobalState() {
+   if (curl_global_init(CURL_GLOBAL_DEFAULT) != CURLE_OK) {
+      throw std::runtime_error(
+          "[CURLWebClient] Failed to initialize libcurl globally");
+   }
+}
+
+CurlGlobalState::~CurlGlobalState() { curl_global_cleanup(); }
+
+namespace {
+// curl write callback that appends response data into a std::string
+size_t WriteCallbackString(void* contents, size_t size, size_t nmemb,
+                           void* userp) {
+   size_t realsize = size * nmemb;
+   auto* s = static_cast<std::string*>(userp);
+   s->append(static_cast<char*>(contents), realsize);
+   return realsize;
+}
+
+// curl write callback that writes to a file stream
+size_t WriteCallbackFile(void* contents, size_t size, size_t nmemb,
+                         void* userp) {
+   size_t realsize = size * nmemb;
+   auto* outFile = static_cast<std::ofstream*>(userp);
+   outFile->write(static_cast<char*>(contents), realsize);
+   return realsize;
+}
+
+// RAII wrapper for CURL handle using unique_ptr
+using CurlHandle = std::unique_ptr<CURL, decltype(&curl_easy_cleanup)>;
+
+CurlHandle create_handle() {
+   CURL* handle = curl_easy_init();
+   if (!handle) {
+      throw std::runtime_error(
+          "[CURLWebClient] Failed to initialize libcurl handle");
+   }
+   return CurlHandle(handle, &curl_easy_cleanup);
+}
+
+void set_common_get_options(CURL* curl, const std::string& url,
+                            long connect_timeout, long total_timeout) {
+   curl_easy_setopt(curl, CURLOPT_URL, url.c_str());
+   curl_easy_setopt(curl, CURLOPT_USERAGENT, "biergarten-pipeline/0.1.0");
+   curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
+   curl_easy_setopt(curl, CURLOPT_MAXREDIRS, 5L);
+   curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT, connect_timeout);
+   curl_easy_setopt(curl, CURLOPT_TIMEOUT, total_timeout);
+   curl_easy_setopt(curl, CURLOPT_ACCEPT_ENCODING, "gzip");
+}
+}  // namespace
+
+CURLWebClient::CURLWebClient() {}
+
+CURLWebClient::~CURLWebClient() {}
+
+void CURLWebClient::DownloadToFile(const std::string& url,
+                                   const std::string& file_path) {
+   auto curl = create_handle();
+
+   std::ofstream outFile(file_path, std::ios::binary);
+   if (!outFile.is_open()) {
+      throw std::runtime_error(
+          "[CURLWebClient] Cannot open file for writing: " + file_path);
+   }
+
+   set_common_get_options(curl.get(), url, 30L, 300L);
+   curl_easy_setopt(curl.get(), CURLOPT_WRITEFUNCTION, WriteCallbackFile);
+   curl_easy_setopt(curl.get(), CURLOPT_WRITEDATA,
+                    static_cast<void*>(&outFile));
+
+   CURLcode res = curl_easy_perform(curl.get());
+   outFile.close();
+
+   if (res != CURLE_OK) {
+      std::remove(file_path.c_str());
+      std::string error = std::string("[CURLWebClient] Download failed: ") +
+                          curl_easy_strerror(res);
+      throw std::runtime_error(error);
+   }
+
+   long httpCode = 0;
+   curl_easy_getinfo(curl.get(), CURLINFO_RESPONSE_CODE, &httpCode);
+
+   if (httpCode != 200) {
+      std::remove(file_path.c_str());
+      std::stringstream ss;
+      ss << "[CURLWebClient] HTTP error " << httpCode << " for URL " << url;
+      throw std::runtime_error(ss.str());
+   }
+}
+
+std::string CURLWebClient::Get(const std::string& url) {
+   auto curl = create_handle();
+
+   std::string response_string;
+   set_common_get_options(curl.get(), url, 10L, 20L);
+   curl_easy_setopt(curl.get(), CURLOPT_WRITEFUNCTION, WriteCallbackString);
+   curl_easy_setopt(curl.get(), CURLOPT_WRITEDATA, &response_string);
+
+   CURLcode res = curl_easy_perform(curl.get());
+
+   if (res != CURLE_OK) {
+      std::string error =
+          std::string("[CURLWebClient] GET failed: ") + curl_easy_strerror(res);
+      throw std::runtime_error(error);
+   }
+
+   long httpCode = 0;
+   curl_easy_getinfo(curl.get(), CURLINFO_RESPONSE_CODE, &httpCode);
+
+   if (httpCode != 200) {
+      std::stringstream ss;
+      ss << "[CURLWebClient] HTTP error " << httpCode << " for URL " << url;
+      throw std::runtime_error(ss.str());
+   }
+
+   return response_string;
+}
+
+std::string CURLWebClient::UrlEncode(const std::string& value) {
+   // A NULL handle is fine for UTF-8 encoding according to libcurl docs.
+   char* output = curl_easy_escape(nullptr, value.c_str(), 0);
+
+   if (output) {
+      std::string result(output);
+      curl_free(output);
+      return result;
+   }
+   throw std::runtime_error("[CURLWebClient] curl_easy_escape failed");
+}

pipeline/src/wikipedia/wikipedia_service.cpp

@@ -0,0 +1,89 @@
+#include "wikipedia/wikipedia_service.h"
+
+#include <spdlog/spdlog.h>
+
+#include <boost/json.hpp>
+
+WikipediaService::WikipediaService(std::shared_ptr<WebClient> client)
+    : client_(std::move(client)) {}
+
+std::string WikipediaService::FetchExtract(std::string_view query) {
+   const std::string encoded = client_->UrlEncode(std::string(query));
+   const std::string url =
+       "https://en.wikipedia.org/w/api.php?action=query&titles=" + encoded +
+       "&prop=extracts&explaintext=1&format=json";
+
+   const std::string body = client_->Get(url);
+
+   boost::system::error_code ec;
+   boost::json::value doc = boost::json::parse(body, ec);
+
+   if (!ec && doc.is_object()) {
+      try {
+         auto& pages = doc.at("query").at("pages").get_object();
+         if (!pages.empty()) {
+            auto& page = pages.begin()->value().get_object();
+            if (page.contains("extract") && page.at("extract").is_string()) {
+               std::string extract(page.at("extract").as_string().c_str());
+               spdlog::debug("WikipediaService fetched {} chars for '{}'",
+                             extract.size(), query);
+               return extract;
+            }
+         }
+      } catch (const std::exception& e) {
+         spdlog::warn(
+             "WikipediaService: failed to parse response structure for '{}': "
+             "{}",
+             query, e.what());
+         return {};
+      }
+   } else if (ec) {
+      spdlog::warn("WikipediaService: JSON parse error for '{}': {}", query,
+                   ec.message());
+   }
+
+   return {};
+}
+
+std::string WikipediaService::GetSummary(std::string_view city,
+                                         std::string_view country) {
+   const std::string key = std::string(city) + "|" + std::string(country);
+   const auto cacheIt = cache_.find(key);
+   if (cacheIt != cache_.end()) {
+      return cacheIt->second;
+   }
+
+   std::string result;
+
+   if (!client_) {
+      cache_.emplace(key, result);
+      return result;
+   }
+
+   std::string regionQuery(city);
+   if (!country.empty()) {
+      regionQuery += ", ";
+      regionQuery += country;
+   }
+
+   const std::string beerQuery = "beer in " + std::string(country);
+
+   try {
+      const std::string regionExtract = FetchExtract(regionQuery);
+      const std::string beerExtract = FetchExtract(beerQuery);
+
+      if (!regionExtract.empty()) {
+         result += regionExtract;
+      }
+      if (!beerExtract.empty()) {
+         if (!result.empty()) result += "\n\n";
+         result += beerExtract;
+      }
+   } catch (const std::runtime_error& e) {
+      spdlog::debug("WikipediaService lookup failed for '{}': {}", regionQuery,
+                    e.what());
+   }
+
+   cache_.emplace(key, result);
+   return result;
+}