I teach machines to read boring PDFs so humans don’t have to — mostly with RAG, Graph-RAG, and cross-encoder reranking.
If there’s a hallucination, I treat it as a bug, not a personality trait — hello Precision@K, NDCG, latency budgets.
I build pipelines that survive real users: FAISS/pgvector, embeddings, normalization, drift checks, sub-second APIs.
Big fan of graphs: turning repos, standards, and messy docs into knowledge graphs + multi-hop reasoning.
Production over demos: FastAPI, Docker, Kubernetes, MLflow, LangChain/LangGraph, observability included.
