At Sonata Software, I work on agentic AI platforms used by enterprise customers. I’m one of the core developers on a LangGraph orchestration engine that runs more than 1,000 agent executions a month.
I wrote supervisor and handoff logic for delegating work across specialist agents, including state checkpointing and retries. I also connected MCP and A2A so agents can share tools and communicate through standard protocols.
For Sonata Software’s multi-tenant AI agent delivery platform, I designed a 3-tier RAG pipeline using document, section, and chunk retrieval. On internal evaluation sets, it improved retrieval precision by about 40% compared with flat chunking.
I’ve also built model-provider adapters, asynchronous evaluation with DeepEval and Ragas, and tenant isolation using PostgreSQL row-level security and JWT-scoped FastAPI endpoints. My work spans services and data schemas through evaluation, tracing, and monitoring.

