I study computer science and physics at Cal State Long Beach, with a minor in biology. I'm a BUILD Scholar working toward MD/PhD training.
I direct AI agents across more than software: running a home estate of several machines, producing research reports, disk and ops cleanup, and building this site. I set the bar and decide what ships.
My research experience spans gait and rehabilitation, behavioral neurobiology, protein molecular dynamics, and collaborative surgical-ergonomics work. I'm interested in measurement, reproducible analysis and scientific software.
Checks come before claims. I keep runnable examples, source and explicit limits with the work. Research contributions are collaborative; synthetic demos are not evidence of clinical validity, learning gains or savings on someone else's data.
| Project | What you can inspect |
|---|---|
| RecallWeave | Offline learning toolkit with visible Bayesian estimates, lesson authoring and interactive labs; no measured learning-gain claim. |
| ReturnBy | Local-first return deadlines and calendar reminders; review extracted fields and merchant rules. |
| TowerOps | Synthetic airspace workbench with approval and readback gates; not operational ATC. |
| Aetherline: Relic Run | Hamon Engine browser prototype; WebGPU required, private engine source, no browser audio. |
| Multi-agent Fan-out Engine | Sanitized reference core for dependency planning and dispatch reconciliation; not a production-capacity benchmark. |
| ShadeWindow | Submitted historical heat-index replay demo; not a safety forecast. |
- CaptureSuite — Windows-first multimodal research capture: a daemon, versioned worker plugins and sealed session packages, with an offline QC demo on synthetic EMG/IMU data.
- CanvasPilot — An MCP server and CLI for Canvas LMS that rides the browser session you already have, for schools that disable personal access tokens.
- workflow-checks — Three small Python checks for back-office paperwork (utility bills, freight detention, PT authorizations), run on synthetic data with file:row evidence.
- Multi-agent Fan-out Engine — A public-safe reference core for dependency plans, SQLite dispatch custody and reconcile-not-replay. Local adapter tests, not claims about production capacity.
Open to research collaborations and scoped pilots: start with a bounded problem, compare with an existing workflow, and measure quality, time, cost and human intervention. No savings or clinical outcomes are implied by a demo.