All projects
Next.js, LangChain, Qdrant, Clerk
AskYourDocs
Natural-language Q&A over your own documents — upload, embed, ask, get cited answers.
Problem
Finding one fact inside a pile of PDFs is slower than it should be.
Why X, not Y
Why Qdrant, not only Postgres
Dedicated vector ops and sub-10ms retrieval without bolting ANN onto an OLTP schema.
Why Clerk
Auth and multi-tenant isolation without rebuilding session plumbing for a docs product.
Why LangChain orchestration
Chunking, retrieval, and answer assembly needed a clear pipeline more than a custom framework.
Flow
- 01Upload docs
- 02Chunk + embed
- 03Retrieve
- 04Answer with cites
If I rebuilt it today
I’d push harder on citation grounding and eval harnesses before adding more model providers.