I build AI systems that take
safety as seriously as capability.

I use AI where it earns its place. Half the fun is figuring out whether it's actually helping or just slowing me down, especially the moment it starts producing slop. Ain't nobody got time to babysit a dumb agent.

• Open-source and local AI: I default to open and local models first, weighing privacy, cost, capability and the job in front of me. Day to day that's LM Studio and Ollama running open weights, Qwen included. At work the stack is closed, mostly OpenAI and Anthropic, which comes with its own tradeoffs.
• Agent loops and graph engineering: I build the harness around agent workflows, explicit state, scoped delegation, handoffs that actually carry context, and memory that doesn't turn into a junk drawer. This is the actual work behind CoworkOS and WorkspaceOS.
• Human-AI product design: I turn messy real-world work into AI products people can understand, trust, and use.
• Evaluation, evidence and safety: I test what a system actually does, keep provenance where it matters, and put review gates around decisions that have real consequences. Same discipline behind chain-of-custody memory and execution guardrails on live trading.
• Product direction and delivery: I move from problem framing and prototype to a working product, with enough technical depth to make good calls along the way.