Projects

Distributed LLM batch processing platform. GCS upload → Eventarc → Cloud Run → dedicated GKE pod per job. Memory-constant streaming — a 10M-prompt job uses the same pod memory as 100.

Production GitHub App — 4 specialist agents (Security, Quality, Blast Radius, LLM) run in parallel on every PR, posting structured inline comments on the diff. 222 passing tests.

AI SaaS converting résumé PDFs into live portfolio sites via structured LLM parsing. 45% latency reduction via Redis caching + S3.

No-code RAG evaluation platform for plug-and-play experimentation across chunking strategies, retrieval modes, and LLM configs. Hybrid retrieval: BM25 + dense + RRF.