How I work

One loop, every project

I use the same six steps for each project. Many AI agents do the build. Gates check the work.

Learn · build · manage

Six steps

Select a lens. Then run one task through the loop.

Read the field. Map competitors. Write a short research note before code.

Research. Read the field. Map competitors. Write a short research note before code.
Plan. Cut the work into small tasks. Each task has one owner and one test.
Build in parallel. One AI agent per git worktree. Each agent works on one branch.
Gate. Build, unit tests, browser smoke tests. Nothing merges if a gate fails.
Ship. Deploy behind a flag. Watch logs. Roll back in one command.
Learn. Write down what broke and why. Feed corrections back into data and docs.

Month by month

I did not start from zero. I started with a base in AI, ML and programming. Then I applied it to harder, real problems each month.

Starting point

Rice University, B.A. Cognitive Science, 2017. I studied how people think and learn, and how those ideas can help machines learn.
A nine-month AI and ML bootcamp, 2019-2020. How models train, how loss and evaluation work, and why models overfit.
About five years in business. Solutions engineer, account manager, and owner of my own business. I learned to scope work with clients and deliver it.

Drag the handle or press Play. The commit numbers come from git logs.

Month

Read the months as text

2025-12: Shipped. A two-sided mobile app with a hosted backend, in two languages. (9 commits)
2026-01: Built. A marketplace where AI agents compete and get paid from escrow. (132 commits)
2026-02: Operated. A fleet of agents on a hardened server, controlled over chat. (203 commits)
2026-03: Applied. My ML training to a real problem: detection models on blueprint tiles. (314 commits)
2026-04: Found. Label quality limited accuracy more than model choice. I changed the data plan. (85 commits)
2026-05: Designed. A human-in-the-loop flow. Each fix becomes training data. (174 commits)
2026-06: Went deeper. OCR and identity matching across near-identical items. (512 commits)
2026-07: Scaled. From solo repos to a team monorepo: web, API and mobile. (877 commits)
2026-08: Hardened. Release gates: build, tests and browser smoke checks. (797 commits)
2026-09: Peak. 2,150+ commits in one month. Many agents, every change gated. (2,156 commits)
2026-10 (to date): Now. Document the work. This site is the result. (95 commits so far)