I build AI agent systems, evaluation tools, and interactive engineering software. My work focuses on making automated work inspectable, reproducible, and useful.
| Project | What it does |
|---|---|
| DSH Operator UI | An operator interface for running agent tasks, inspecting execution evidence, and managing reusable capabilities. |
| FlowRouter | A capability-manifest specification for describing executable work, its inputs and outputs, and evidence of execution. |
| FishBench | Model evaluations covering persona consistency and generated 3D scenes, with recorded results to inspect. |
| INDUSTRIA | An interactive machining simulation exploring cutting force, deflection, stability, and dimensional error. |
| Shop OS — Pillow-Block Proof | A documented manufacturing-agent run with downloadable CAD, advisory process plans, and runnable verification tools. |
RCOS (Recursive Capability Operating System) is the research direction connecting my agent work: execute a task, evaluate the result, and reuse capabilities that meet explicit admission criteria. The public operator UI and FlowRouter manifest specification show parts of that work. Their repositories document what is available today.
I’m interested in the full path from a request to a result: how agents execute work, how we evaluate the output, and how useful capabilities can be reused. The projects above explore that through agent tooling, benchmarks, and engineering simulations.
Each repository documents its own scope, setup, and limitations.