Tech Lead AI at atQor. Around six years taking ML and AI systems from prototype to production: RAG pipelines, agents on Claude Agent SDK and OpenAI Agents SDK, MCP servers, and the tracing and evals that keep them honest after launch.
retrieve()Retrieval that stays grounded: chunking, hybrid search, reranking, and answers you can trace back to a source.
handoff()Production agents on Claude Agent SDK and OpenAI Agents SDK, with clear roles, handoffs and guardrails.
tools/listModel Context Protocol servers that give agents typed, safe access to real systems and data.
judge.score()Langfuse tracing and LLM-as-judge evals, so quality is measured release over release, not guessed.
Built a production agent on Claude Agent SDK and led the team building a second, for a safety data sheet (SDS) management platform. Backed by an LLM-as-judge eval pipeline, with demos to product owners, management and CXOs.
A reusable natural-language-to-SQL template that turns a client's database into a POC-ready demo with only config changes. Intent detection, retrieval-grounded SQL generation, secrets and monitoring built in.
An agent that triages incoming incidents: classifies, summarises and routes them so engineers start with context instead of a raw alert.
A small RAG agent over my case studies, with its trace shown live.