RAX - RAG API

A Retrieval-Augmented Generation API that combines semantic search over PDF documents with LLM-powered answer generation using Neo4j, OpenAI, and FastAPI.

How It Works

1. Ingest — PDFs in /data are semantically chunked, embedded via OpenAI (text-embedding-3-small), and stored in Neo4j as vectors.
2. Query — User questions are embedded and matched against stored chunks using cosine similarity. The top results are fed as context to GPT-4o-mini, which generates an answer with source attribution.

Prerequisites

• Python 3.10+
• uv (package manager)
• Docker & Docker Compose (for Neo4j)
• OpenAI API key

Setup

# Install dependencies
uv sync

# Start Neo4j
docker-compose up -d

# Configure environment
cp .env.example .env  # then add your OPENAI_API_KEY

Usage

Ingest documents

Place PDF files in the /data directory, then run:

from rax.ingest import run
run()

Start the server

python main.py

The API will be available at http://localhost:8000.

Query

curl "http://localhost:8000/ask?q=your+question+here"

Response:

{
  "question": "your question here",
  "answer": "...",
  "sources": ["filename.pdf"]
}

Project Structure

rax/
├── rax/
│   ├── config.py      # Environment and model configuration
│   ├── ingest.py      # PDF loading, chunking, embedding, and storage
│   ├── query.py       # Vector retrieval and LLM answer generation
│   └── server.py      # FastAPI application
├── data/              # PDF documents to index
├── main.py            # Entry point (runs uvicorn)
├── docker-compose.yml # Neo4j service
└── pyproject.toml     # Dependencies and project metadata

Tech Stack

Component	Technology
Vector DB	Neo4j 5.0
Embeddings	OpenAI text-embedding-3-small (1536-dim)
LLM	GPT-4o-mini
API	FastAPI + Uvicorn
Document Processing	LlamaIndex (semantic splitting)

License

MIT