I build strange, useful things with AI.

I’m interested in the parts of AI that become tangible in someone’s actual life: agents that keep context, interactions that preserve human agency, small tools that change a stubborn moment, and evaluation that tells us whether any of it really works.

This profile is a public workshop rather than a finished portfolio. I’d rather leave a shelf empty than fill it with something I wouldn’t stand behind.

An AI study buddy that starts first.

Millie explores task initiation for neurodivergent learners by changing the usual tutoring dynamic: the agent begins a small part of the learner’s real assignment, reaches a specific sticking point, asks for help, then visibly changes its work before handing one manageable action back.

Repository · Live prototype · Devpost

Work contributed upstream lives here, separate from my own projects. The wall refreshes automatically from public PRs.

• Human–Agent interaction — continuity, initiative, shared workspaces, and new interaction primitives
• Persistent personal agents — memory and context that can survive across tools and surfaces
• Agent skills & harnesses — small reusable capabilities instead of monolithic “do everything” agents
• Evaluation — turning vague impressions of agent quality into observable evidence
• Embodied interfaces — what changes when an agent can inhabit a desk, room, device, or physical routine

I’m trying to make this profile a forcing function for finishing things. A project becomes a public artifact when it has:

1. a clear why — one sentence about what changes for a person or agent;
2. a runnable or installable entry point;
3. a real demo or GIF;
4. a README that makes sense without private context;
5. an explicit license.

Forks and upstream work stay in Open-source contributions. Experiments can stay experiments until they’re ready.

_{Building in public, selectively.}