AI · 2026In progress

RAG Knowledge Assistant

Ask questions over your own documents and get answers grounded in them. Files are chunked and embedded into a vector index, and every answer cites the passages it came from.

Most teams keep their knowledge in PDFs, wikis and Slack threads that nobody can search well. This assistant ingests those sources, splits them into overlapping chunks, stores embeddings in MongoDB Atlas Vector Search, and at query time retrieves the most relevant passages before asking Claude to answer strictly from them. Every answer links back to its sources, and a feedback loop records which retrievals were useful so the index and chunking can be tuned over time.

Challenges

Chunking long documents without losing context across chunk boundaries
Keeping answers grounded instead of letting the model fill gaps from memory

Solutions

Overlapping chunks with document and section metadata attached to every embedding
A strict prompt contract: answer only from retrieved passages, and say so when the sources don't cover the question

Next.jsPythonClaudeMongoDB Atlas Vector Search