I build AI and machine learning you can trust, and I write down how.
Project deep dives on building AI, the architecture, the trade offs, and the evaluation behind each decision. Hover a project below to see how it flows.
Automate the workflow. Keep the evaluation visible.
One scikit learn engine behind a web console, a LangGraph agent, and an MCP plugin, now with a grounded AI copilot that reads your data and an agent that plans the run.
Project perspectives
Hover to preview, click to readAutoDS Copilot
A no code data science copilot. It runs the whole ML workflow from a spreadsheet, answers questions from your real data, and cross validates the models to pick the honest winner.
Read the post →SupportIQ
A support agent that answers what it can prove and hands the risky cases to a human before anything is sent.
Read the post →SKU Sentinel and NextMove
Watching demand forecasts drift the way an SRE watches servers, then turning them into order quantities.
Read the post →The unglamorous half of machine learning.
I care about the parts most people skip, the monitoring, the grounding, the human oversight, and the honest evaluation. A model is easy to build and easy to fool, so the work that interests me is the layer that makes a result trustworthy enough to act on.
I am always building toward systems that turn predictions into decisions a business can actually use, and I write each one up here so the thinking is out in the open. If you work on applied machine learning, agents, or trustworthy AI, I would love to talk.