What I’ve built
Each case study follows the same structure: the problem, the design, the numbers, and what went wrong and what I changed. The last section is the one I would read first.
Health AI Central
An independent intelligence hub for the health AI field, run by 40+ agents I designed, and one human decision that nothing publishes without.
Journal of Health AI
A new international, peer-reviewed, gold open-access journal, with a published AI-use policy and a pipeline where every editorial judgment is human.
Red-teaming my own apps
I ship web applications people in health care actually use, so I attack them before anyone else can: audits, injection tests, and fixes with dates.
Hospital-system strategy simulation
A multiplayer simulation for graduate healthcare management courses: teams govern competing hospital systems, advised by LLM division directors, and a deterministic engine keeps score.
The infrastructure underneath
Everything above runs on infrastructure I operate myself: five virtual servers running Docker services (one carries Health AI Central; the others carry other projects, including an LLMOps and penetration-testing platform I use to optimize my generative AI applications and harden my own infrastructure). Underneath them sit a reverse proxy, a self-hosted mail server for five domains, uptime monitoring across every host, and a gateway that routes every model call under a per-agent key. Telemetry captures every API call, so I can go back over any answer, see what it cost, and make the pipelines cheaper and more accurate month over month.
Running these services myself is not frugality; it is the design. Self-hosting makes the security surface mine to harden and mine to red-team. Per-agent keys and call-level telemetry make every action auditable. And putting a gateway between the agents and the models makes scaling a configuration change, not a rewrite.