🏢 Company HR Policy Assistant (RAG Pipeline)

This project is a production-grade Retrieval-Augmented Generation (RAG) system built from scratch in Python. It currently serves as a FastAPI backend, utilizing semantic chunking, OpenAI embeddings, and a local ChromaDB vector store.

🗺️ Codebase Map

Root Directory

• main.py: The entry point. It orchestrates the database setup and runs the FastAPI web server via Uvicorn.
• ROADMAP.md: Tracks project progress and upcoming architectural stages.
• .env: (Git-ignored) Stores the OPENAI_API_KEY.
• requirements.txt: List of dependencies for the uv environment.

app/ (Core Application Logic)

• api/ (Network Layer)

  • rag.py: Contains the FastAPI routes (e.g., POST /api/v1/query) that expose the RAG engine to external clients.

• db/ (Data Contracts)

  • models.py: Uses Pydantic to strictly define internal data (Document, Chunk, RetrievedChunk) and external network payloads (QueryRequest, QueryResponse).

• ingestion/ (Data Processing)

  • loader.py: Reads raw .md files from the docs/ folder.
  • chunker.py: The "Smart Semantic Chunker." Uses regex to split text safely at paragraph and sentence boundaries.

• embeddings/ (Math & Vectors)

  • embedder.py: Talks to OpenAI's text-embedding-3-small API to convert text chunks into 1536-dimensional vectors.

• vectorstore/ (Database Operations)

  • vector_db.py: The isolated ChromaDB client. Handles saving (upsert_chunks) and searching (query_db) the vector graph.

• rag/ (LLM Interaction)

  • generator.py: Constructs the grounded system prompt and interacts with gpt-4o-mini to generate cited answers.

Other Folders

• docs/: The raw Markdown files (HR policies) that the system reads from.
• tests/: Contains test_core.py to automatically verify data models and DB connections using pytest.