I teach teams to work with AI and build tools around tasks I know. These examples show the decisions, the working artifacts, and what I learned along the way.
Public workshops, personal builds, and hackathon prototypes. Each is labeled so you can judge the evidence in context.
In Build Your First Agent Loop, participants used Loop Lab to turn a recurring task into a plan: what goes in, what the agent may change, how to check the result, and when a person steps in.
What this shows: how I make a technical idea concrete enough to work through with a partner.
Read the workshop case study →A guided sales-workflow prototype that turns deal context into a visible prompt. The user can inspect the text, copy it, or hand it off to another tool.
What this shows: how I draw on sales experience to give a recurring task a clearer starting point.
Read the Prompt Runner case study →
A useful agent needs a result someone can inspect. This prototype keeps the reviewer packet beside the sources and code-check receipt.
A multi-agent peer-review prototype with a shared evidence board and a sandbox check for the paper's own code. It won the scientific research track and best overall at the May 2026 AG2 hackathon.
Read the RefereeOS case study →
Shared evidence board Claims and concerns stay visible beside their cited source.
Sandbox receipt The reviewer packet keeps the code check and its result inspectable.
Eight written build case studies across this page, plus the workshop and a private prototype available by conversation.
Tell me what repeats, where it gets stuck, and who needs the result. I'll help you decide whether a workshop, a small build, or a second opinion makes sense.
Tell me about it →