Safee Rehman

At MATZ Solutions, I lead AI work for HalalOne, building a multi-node LangGraph agent that answers certification queries across a catalog of more than 200,000 records. I designed retrieval across keyword, semantic, and web search, plus WebSocket streaming that keeps in-flight responses resilient through client disconnects.

I build agent systems end to end: translating ambiguous needs into workflows, tool calling, structured outputs, LLM business logic, and deployment. For Tadabbur, I created a multi-tool retrieval-augmented agent over 6,200+ tafsir and Asbab al Nuzul passages, with sub-agent routing for retrieval, story generation, and image illustration.

I care about controlled, reliable agent behavior. I've built LangSmith and custom evaluation harnesses for intent classification, tool trajectories, retrieval relevance, structured outputs, and hallucination rates, and I also developed a real-time voice agent to autonomously run London Halal Forum 2025.