I work on better ways to do research with AI: clear questions, checkable results, and corrections that don't get lost between sessions.
Research portfolio · Research résumé
- Five Graffiti³ counterexample families — exact family arguments with bounded computational corroboration.
- Semiorder projection bounds — a largest-index zonotope decomposition; unrestricted parametric closure remains open.
- Minimal-counterexample degree bounds — a scoped theorem, not a solution of the Erdős–Gyárfás conjecture.
- A 104-step totient certificate —
F(400000287233629) = 104; not a maximum, record, or unboundedness claim.
The three papers are public preprints with internal, method-diverse checks; external specialist review is pending in their recorded release state.
I choose the problems, frame the tasks, specify verification, challenge claims, and direct corrections and releases. AI systems substantially assist with exploration, proofs, code, computation, checking, and writing. I don't claim to have independently written every proof or implementation.
The wider ETI program studies how human–AI workflows preserve evidence and handle corrections. Comparative efficacy is still untested.
I'm interested in AI evaluation and research-workflow roles. Get in touch.