2026 · AI / ML / Agentic AI
NASA Intelligence Chat
A retrieval-augmented chat over NASA mission documents with source citations and RAGAS evaluation.
- NASA mission documents
- ChromaDB retrieval
- Cited answers
Complete portfolio projectGitHub ↗
Problem
NASA mission transcripts and technical reports are long and dense; plain keyword search can't answer questions about them directly.
My role
Author of the repository.
Approach
- Ingest 12 NASA documents into a ChromaDB vector store using OpenAI embeddings.
- Retrieve top-k chunks by cosine similarity, then generate answers with [Source N] citations.
- Score each response with RAGAS (response relevancy and faithfulness); Streamlit UI with mission filtering.
Results
- Evaluated on 18 questions; the README reports mean response relevancy ≈ 0.89 and faithfulness ≈ 0.86.
Stack
- Python
- OpenAI
- ChromaDB
- RAGAS
- Streamlit
- pytest