Document Retrieval Agent App

This app is a document-grounded AI chatbot that answers queries from:

• A folder of 5 .txt FAQ documents (for FAQ search)
• A .csv file of monthly sales (columns: Date, Product, Sales; for sales analytics)

Features

• LangChain agent with Supervisor for tool routing (FAQ and sales)
• FAQ agent always answers directly from the FAQ documentation, never from general LLM knowledge
• FAQ search using OpenAI embeddings and vector search
• Sales data Q&A using SQL over a SQLite database (automatically loaded from CSV)
• FastAPI backend with streaming responses (SSE)
• Streamlit chat UI for interactive chat (port 8501)
• Uses OpenAI API for both LLM and embeddings (set your OPENAI_API_KEY)
• All test files are in the test/ directory

Setup (Local)

1. Install dependencies:

   pip install -r requirements.txt
   pip install streamlit sseclient

2. Set your OpenAI API key in your environment:

   export OPENAI_API_KEY=sk-...

3. Place your 5 FAQ .txt files in data/faqs/.
4. Place your sales data in data/sales.csv (columns: Date, Product, Sales).
5. Run the backend:

   uvicorn app.main:app --reload

6. Run the Streamlit chat UI:

   streamlit run app/streamlit_chat.py

Docker Compose Deployment

1. Add your OPENAI_API_KEY to a .env file in the project root:

   OPENAI_API_KEY=sk-...

2. Build and start both backend and UI:

   docker-compose up --build

3. Access:
   • FastAPI backend: http://localhost:8000
   • Streamlit chat UI: http://localhost:8501
   • The Streamlit UI is preconfigured to use the backend at http://app:8000/chat (Docker Compose networking).

Usage

• Send POST requests to /chat with { "message": "your question" }.
• Responses are streamed back as text chunks (SSE).
• The agent will automatically route questions to the FAQ or sales SQL agent as appropriate.
• FAQ answers are always grounded in your documentation. If no answer is found, the agent will say so.

Testing

• All test files are in the test/ directory.
• To run a test manually:

  python test/test_faq_agent.py
  python test/test_sales_agent.py
  python test/test_faq_tool.py
  python test/test_fastapi_chat.py

• The FastAPI chat endpoint can be tested with test/test_fastapi_chat.py (streams SSE responses).

---