FastAPI Copilot Doc — RAG for FastAPI Documentation

1. Project Description

FastAPI Copilot Doc is an offline Retrieval-Augmented Generation (RAG) system that indexes FastAPI markdown documentation and answers developer questions by grounding responses in the indexed docs.

2. Project features

• Document loading & chunking via src.indexing.create_chunks
• Indexing pipeline (walk markdowns → chunks → store) via src.indexing.initiate_indexing_process
• FAISS vector store for retrieval (index saved to ./faiss_index) via src.indexing.vector_store and src.vector_store.get_vector_store
• HuggingFace embeddings (sentence-transformers/all-MiniLM-L6-v2) used in src.indexing / src.vector_store
• Agent with middleware that injects retrieval context into prompts: src.agent.agent + src.middleware.prompt_with_context
• URL verification tool for citations: src.tools.validate_fastapi_doc_url
• Interactive CLI QA bot: src.rag.initiate_qa_bot
• Local loader test: src.test_load_and_chunking.py

3. LangChain concepts used

• Retrieval-Augmented Generation (RAG) with embeddings + FAISS
• Vector store operations and similarity search via src.vector_store.retrieve_similar_documents
• Agents, middleware, and tools (see src.agent.agent, src.middleware.prompt_with_context, src.tools.validate_fastapi_doc_url)
• Middleware for summarization & retry: SummarizationMiddleware, ModelRetryMiddleware (configured in src.agent)
• Streaming responses via the agent (used in src.rag)

4. Steps to run this project locally

Prerequisites

• Python version (see .python-version)
• Copy .env.example → .env and set required provider keys: .env.example
• Project deps are defined in pyproject.toml

Install & run (using uv package manager as provided)

1. Install dependencies:

uv install

2. Index the docs (creates / updates ./faiss_index):

uv run python -m src.indexing

This runs src.indexing.initiate_indexing_process which loads markdown files and calls src.indexing.vector_store.

3. Run the interactive QA bot:

uv run python -m src.rag

Notes:

• If you prefer not to use uv, you can use your system python (e.g., python -m src.indexing).
• Ensure environment variables for your LLM provider are set in .env before running the agent.

5. LangSmith compatibility

• The project uses LangChain agents and middleware (src.agent, src.middleware), so it is straightforward to plug into LangSmith for monitoring and run management.
• To use LangSmith:
  • Configure a LangSmith-compatible LLM or client in src.agent.agent.
  • Set LANGSMITH_API_KEY (and other LangSmith settings) in .env.
  • Optionally route agent run logs / traces to LangSmith for inspection and debugging.
• Because the project splits retrieval (src.vector_store) and agent logic (src.agent + src.middleware), swapping to a LangSmith-orchestrated workflow is localized and minimal.

---

Files of interest:

• src/indexing.py — loader, chunking, and indexing
• src/vector_store.py — vector store access & retrieval
• src/middleware.py — prompt context injection
• src/agent.py — agent setup & middleware
• src/tools.py — URL validation tool
• src/rag.py — interactive CLI QA bot
• pyproject.toml | .env.example