I build for the physical world.
I work on autonomous machines, new infrastructure, and experiments that test whether ambitious ideas can work.
Play.
Tools and experiments that made an idea concrete.
- Unseeded solver searchEvolves six-instruction solver programs from random starts. In a small study, two of three winners solved all 48 held-out systems.
- tywrapGenerates typed TypeScript wrappers for Python libraries across Node, Bun, Deno, and the browser.
- torch-dfoDerivative-free optimization on PyTorch tensors, written for GPU workloads.
- zergraphAn in-memory Rust property graph. Snapshots and deltas merge changes from independent writers.
- tw-layout-lintChecks layout-only Tailwind v4 classes emitted by coding agents and returns stable repair codes.
- WeftTyped Gleam layout constructors that compile to deterministic, deduplicated CSS.
- Agent Plugin TranscoderBuilds Claude and Codex packages from one source repository.
- The Pundits Tool-Chain archiveRecovered and documented Kevin S. Braunsdorf's Unix administration tools.
Notes, musings, and dreams.
Short notes
- Cheap matters only after good enough.When quality and cost share one score, how do we keep a cheap failure from winning? Reject anything below the quality bar, then compare cost among the results that pass.
- A fix has to reach the weak spot.An equation solver can struggle because one part of the problem is harder than the rest, but what if a broad correction misses it? Identify the weak part and correct it directly.
- Unused work can still enforce a rule.A program may calculate a value only to reject it when it breaks a rule, so can we safely delete that unused calculation? Only when we can prove its check could never fail.
- A debugging step should not change the result.Simulations use a fixed seed so a run can be repeated, but how can we add a debugging preview or test sample without changing every random choice that follows? Give each phase its own source of random numbers.
What I took with me.
- Infodesk
Chief Product Officer (CPTO)
Leading product and engineering taught me that strategy has to survive contact with the company carrying it out. Technical ambition is useful only when the organization can finance it and deliver it.
- App Annie, later data.ai
Head of Data.ai Labs
We built machine-learning systems at very large scale. The model is only a small part of the job; data pipelines and repeatable operations determine whether it becomes a product.
- Tangoe
Director of Product Management & Operations
At Tangoe I learned how distressed turnarounds narrow choices when time and cash run short. Private-equity rollups taught me that integration creates value only when it preserves what made the acquired companies work.
- FedEx
Senior Engineering Advisor
It wasn't like the Tom Hanks movie. A huge organization taught me that politics and inertia are engineering constraints. A sound idea goes nowhere unless people can adopt it without stopping the operation.
Education
- MS, FinanceIndiana University Bloomington, Kelley School of Business
- Certificate of Applied MathematicsIndiana University East · Summa Cum Laude
- Bachelor's, Information Systems & MarketingUniversity of Memphis · Magna Cum Laude
Say hello.
Questions, ideas, or a hard problem worth solving.