A full-stack Retrieval-Augmented Generation (RAG) system for querying NASA bioscience research. Built for the NASA Space Apps 2025 NYC Hackathon, where it won Best Cloudflair Implementation award for using the entirety of cf stack in its approach to document processing and AI-powered knowledge retrieval.

The project connects a serverless backend (Cloudflare Workers) that ingests PDFs and runs semantic search with a space-themed React frontend where users can chat with an AI that answers from those documents.

Astro is a document-to-query system: you ingest PDFs (e.g. NASA reports or research papers), and the system turns them into a searchable knowledge base. Users ask questions in natural language and get answers grounded in those documents, with source citations and optional web search for broader context.

The project was created for the NASA NYC Hackathon with two goals:

1. Document intelligence — Turn dense PDFs (text, tables, images) into structured, searchable content using AI.
2. Accessible research — Let anyone query NASA bioscience research without manually reading long documents.

The result is a cloud-native RAG stack (Cloudflare D1, Vectorize, Workers, and Google Gemini) that is scalable and runs at the edge.

1. Ingestion: PDF URLs are sent to the backend. Text is extracted, tables and images are processed, and content is chunked and embedded.
2. Storage: Chunks go into Cloudflare D1; embeddings go into Vectorize.
3. Query: User questions are embedded, matched to chunks via vector search, and answered by Gemini with source attribution.
4. Interface: The frontend provides a chat UI with voice input, hybrid search mode, and relevance indicators for sources.

cf/
├── README.md                    # This file — project overview and navigation
│
├── cf_ai_astro-rag/             # Backend: RAG API (Cloudflare Workers)
│   ├── src/
│   │   └── index.js             # Worker entry point, routing, ingest/query handlers
│   ├── queryHandling.js         # RAG query logic (embedding, vector search, response)
│   ├── queryHandling.py         # Python reference implementation
│   ├── test/
│   │   └── index.spec.js        # Test suite
│   ├── wrangler.jsonc           # Cloudflare Workers config (D1, Vectorize, env)
│   └── package.json             # Dependencies (pnpm)
│
└── Astro/                       # Frontend: chat UI
    ├── README.md                # Frontend setup and config
    └── astro-nasa-chat/         # React + Vite app
        ├── src/
        │   ├── App.jsx          # Main app, chat UI, API integration
        │   ├── main.jsx         # Entry point
        │   └── index.css        # Tailwind styles
        ├── index.html
        ├── package.json         # Dependencies (npm)
        └── vite.config.js       # Vite configuration

cd cf_ai_astro-rag
pnpm install
pnpm run dev          # Local development
npx wrangler deploy   # Deploy to Cloudflare

See cf_ai_astro-rag/README.md for D1, Vectorize, and env setup.

cd Astro/astro-nasa-chat
npm install
npm start

Open http://localhost:3000. Configure the API endpoint in App.jsx to point at your deployed backend.

Winner of the NASA NYC Hackathon — Recognized for combining document processing, vector search, and AI chat into an accessible research tool.

MIT — see the LICENSE file in the repository.