DevPulse50,000 documents as one searchable mind.
The problem
Engineering teams write more than they can read. Context lives buried in docs, postmortems and Slack threads. Keyword search is shallow and LLMs hallucinate without grounding.
What I built
A retrieval-augmented platform fusing dense embeddings (pgvector) with sparse BM25 via Reciprocal Rank Fusion, with citation-enforced generation and refusal on low confidence, so every answer returns its source.
+23%Recall@5 over a pure-dense baseline on a 50K-document corpus
2services: Spring Boot API tier + FastAPI AI worker, scaling independently, bridged by Redis Pub/Sub
