Skip to content

AI & ML · Research prototype

DocPilot RAG

Documentation Q&A combining keyword and dense retrieval, cross-encoder reranking, and source citations.

PythonLangGraphQdrantFastAPI
Explore repository
DocPilot RAG project cover

The project

A FastAPI documentation assistant with a React interface. It combines BM25 and Qdrant dense retrieval through reciprocal rank fusion, with optional cross-encoder reranking. LangGraph routes questions through retrieval, a bounded query rewrite, and generation; SQLite caches embeddings and complete responses. The repository includes retrieval and generation ablations, but its evaluation questions are synthetic and not manually reviewed. Cloud deployment files are provided as configuration, without a verified hosted application.

Repository activity ·

Inside the build

  • 01BM25 and Qdrant hybrid retrieval with reciprocal rank fusion
  • 02Optional cross-encoder reranking before selecting source passages
  • 03LangGraph routing, one query-rewrite retry, and numbered source citations
  • 04Persistent embedding and response caches plus evaluation scripts

Explore more

ReliefStream

Centralizes flood rescue requests with AI extraction, deduplication, a map, and volunteer coordination.

AeroOps ML

An aviation ML pipeline for flight-delay prediction and engine remaining useful life, with data-quality checks and model serving.