Thesis RAG Chat

A RAG (Retrieval-Augmented Generation) app that lets you chat with a PDF thesis. It ingests the document into a vector database and uses semantic search + an LLM to answer questions grounded in the actual content.

Tech Stack

• FastAPI — API server
• LangChain — RAG pipeline (LCEL)
• ChromaDB — vector store
• OpenAI — embeddings (text-embedding-3-small) + chat (gpt-4o-mini)
• pymupdf4llm — PDF to Markdown conversion
• PostgreSQL — query logging
• Docker Compose — containerized deployment

How It Works

1. Ingestion: PDF is converted to Markdown, split by document headers (chapters/sections), then sub-split into ~1500-char chunks. Each chunk is embedded and stored in ChromaDB with structural metadata.

2. Query: User question is embedded, top-k similar chunks are retrieved, and the LLM generates an answer citing specific chapters and pages.

Project Structure

app/
  main.py          # FastAPI entry point
  api.py           # /query, /health, /history endpoints
  schemas.py       # Pydantic models
  static/          # Chat UI (HTML + CSS + JS)
rag/
  ingest.py        # PDF chunking (hybrid structural + recursive)
  vectorstore.py   # ChromaDB setup
  rag_chain.py     # Retrieval + LLM chain
  prompts.py       # Prompt template
db/
  database.py      # SQLAlchemy engine
  models.py        # QueryLog model

Setup

Docker (recommended)

# 1. Create .env with your OpenAI key
cp .env.example .env

# 2. Place your thesis PDF
cp your_thesis.pdf thesis/master_thesis.pdf

# 3. Start services
docker compose up --build

# 4. Ingest the PDF (once)
docker compose exec api python -c "from rag.ingest import ingest_pdf; print(ingest_pdf('thesis/master_thesis.pdf'))"

# 5. Open http://localhost:8000

Local

pip install -r requirements.txt
# Set up PostgreSQL and update DATABASE_URL in .env
uvicorn app.main:app --reload

API

Endpoint	Method	Description
/	GET	Chat UI
/query	POST	{"question": "..."} → {"answer": "...", "sources": [...]}
/health	GET	Health check
/history	GET	Last 20 queries