Production proof
At DataRobot, I designed Global MCP around a small semantic search/execute surface for a roughly 4MB OpenAPI REST surface, letting agents reach full SDK-scale capability without wasting every turn on tool catalog context.
I also shipped secure sandboxed code execution through per-call Kubernetes Jobs with egress controls, strict security posture, and production recovery paths. That is the kind of work I want the site to make obvious first: agent systems, platform constraints, safety boundaries, and shipping discipline.
Background
At DataRobot, my recent work has focused on next-generation enterprise AI platform capabilities: MCP-based architectures, OpenAPI-driven tool interoperability, search and execution workflows, secure sandboxed code execution, human-in-the-loop controls, skills/tool ecosystems, agent templates, and enterprise-ready OpenClaw enablement.
Earlier, I led trust and explainability work, contributed to core AutoML systems, helped build Real Python, led scientific software development at Harvard Medical School, and built a software consultancy in Malaysia serving enterprise customers across aviation, banking, energy, automotive, retail, and finance.
Current focus
I am focused on agent platforms that can survive production pressure: tool discovery, execution controls, evals, approval flows, observability, rollback paths, and interfaces that make powerful systems understandable.