Overview
An internal knowledge platform: documents are ingested, chunked, embedded and retrieved with a hybrid pipeline, then answered by an agent. The value is the pipeline, not the chat box.
Role
Architect & full-stack engineer
Stack
- Next.js
- ASP.NET Core .NET 9
- Python FastAPI
- Postgres
- Weaviate
- Redis
- R2
Highlights
- 01Document pipeline: extract → chunk → embed → store, run in the background, never in the upload request
- 02Hybrid retrieval: BM25 + vector + rerank, results normalised and cited before reaching the UI
- 03Agent workflow + MCP server surface
- 04Strict multi-tenant isolation — tenant_id everywhere; cross-tenant access returns 404 (anti-enumeration)
Outcomes
3
stacks orchestratedBM25+kNN
hybrid retrieval