Thirteen ventures built end to end since 2015. Ten are live right now, running on LLM APIs, agentic workflows, and evals in production.
See the build record681 of my own working sessions, read back as operating behavior: how I decompose a problem, how I handle being wrong, how fast I close a loop.
Read the analysisI pay my own inference bills. Cost, latency, and eval quality trade against each other in production, and I have made those calls with my own money on the line.
Ten live right now, two in build. Every one went from the first line of code to real users without a team. The AI products run on frontier model APIs, agentic workflows, and evals in production, not in a demo.
Software with AI at the core, learning, chat, vision, gameplay.
Physical brands shipped from Detroit.

Ten scents. Wash, lotion, soap. From Detroit.
Snap it, print it, hold it. Custom figurines made in Detroit.
Focused products with devoted users, audio, marketplaces, game companions.
Brands I built and still keep on the internet. I am not adding features to these. They stay up because they work.
Deep V is the exception, shut down and no longer online. Listed anyway, because the work happened.
Photograph a small thing worth remembering. Get it back as an object you can keep.
Quest tracker for all 103 quests, plus a 628-item database.
AI NPC engine and SDK for game developers. Unity, Unreal, Godot.
Find local DJs and hear their sound, then request a direct intro.
AI play calling for Madden 26 Ultimate Team.
Premium deep V-neck tees, long sleeves, undershirts. Shut down, store closed.
Every square is a day I put code into one of these products, auto-scanned from my machine twice a day.
No playbook, no team, no clean brief. I build the structure as I go, and I would rather ship one thing that serves ten customers than ten things that serve one each.
End to end ownership from the first conversation to the thing running in production. Nothing sits waiting on a handoff that never comes.
When I am wrong I say so early, out loud, and move. Nearly three years of my own working sessions are on the record for anyone who wants to check that.
Anyone can get a model to do something impressive once. The work is everything after: the evals, the cost per call, the latency budget, the failure mode nobody planned for, the thing that has to keep working on a Tuesday when nobody is watching. I have walked that whole distance on my own products. It is why I can tell a customer the truth about what comes next.
Building something and want to compare notes, partner up, or just connect? This is the fastest way to reach me. Same address if you are only looking through the portfolio and something caught your eye. I read and answer everything myself.
Email me →jack@dotcomjack.com