# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Development Commands

**Dependencies & Setup:**
- `uv sync` - Install dependencies using uv (recommended)  
- `cp .env.example .env` - Setup environment variables, then edit `.env` with your API keys

**Running the Application:**
- `streamlit run src/streamlit_app.py` - Start the Streamlit web application (available at http://localhost:8501)
- `uv run streamlit run src/streamlit_app.py` - Alternative using uv

**Testing:**
- `pytest -q` - Run tests quietly
- `pytest src/` - Run tests for specific directory
- `pytest -v` - Run tests with verbose output

**Development:**
- `uv add package-name` - Add new dependencies
- `uv run python -m streamlit run src/streamlit_app.py` - Alternative run command

## Architecture Overview

This is a Streamlit-based AI resume and cover letter generator that analyzes job descriptions against user background to create tailored documents.

**Core Architecture:**
- **Frontend**: Streamlit app (`src/streamlit_app.py`) with sidebar inputs and main preview area
- **Services Layer**: Business logic split into focused services
  - `analyse_service.py` - JD analysis and user feedback refinement
  - `generation_service.py` - Resume/cover letter generation  
  - `pdf_service.py` - PDF export functionality
  - `llm_service.py` - LLM interaction wrapper
- **LLM Layer**: LiteLLM integration for multi-provider support
  - `litellm_client.py` - API client wrapper
  - `prompt_templates.py` - Template management

**Data Flow:**
1. User inputs JD + background → Analysis service → LLM analysis
2. Analysis summary → Generation service → Resume (MD) + Cover letter (text)  
3. Generated content → PDF service → Exportable PDFs

**LLM Integration:**
- Uses LiteLLM for multi-provider support (OpenAI, Azure, Gemini)
- Responses parsed as JSON for structured analysis data
- Templates centralized in `llm/prompt_templates.py`

**PDF Generation:**
- WeasyPrint for Markdown → PDF conversion
- Exports to timestamped files in `exports/` directory

**Key Dependencies:**
- `streamlit` - Web UI framework
- `litellm` - Multi-provider LLM client  
- `weasyprint` - PDF generation
- `pydantic` - Data validation

**Project Structure:**
```
src/
├── streamlit_app.py          # Main application entry point
├── services/                 # Business logic layer
│   ├── analyse_service.py    # Job description analysis
│   ├── generation_service.py # Document generation
│   ├── llm_service.py       # LLM interaction wrapper  
│   └── pdf_service.py       # PDF export functionality
└── llm/                     # LLM integration
    ├── litellm_client.py    # Multi-provider API client
    └── prompt_templates.py  # Centralized prompt management
```

**Important Notes:**
- Application uses Streamlit session state for workflow persistence
- Environment variables must be configured in `.env` before running
- Supports OpenAI, Azure OpenAI, and Google Gemini via LiteLLM
- PDF exports are saved to `exports/` directory with timestamps