proof of learning
Building with AI as the workforce
What one builder learned to do in an AI-native engineering program, shown as running software instead of a slide deck. Every claim below links to something you can open, run, or independently verify — a live product, a public repository, or a block explorer.
read this first — from the builder
I'm first a reasoning-and-logic developer. If the look isn't there, here are my words: judge me on the logic and the complexity of the project before you judge the way my button looks. I build fullstack, leaning backend — which is where fintech lies. Everything below is that argument, with receipts.
One monorepo, many products
one monorepo · shared packages · one deploy pipeline
Booking, marketplace payments, inventory + POS, intelligence, and onchain — a family of products built from a single monorepo with shared packages, one deploy pipeline (canary boot + automatic rollback), and a shared hosted sign-in.
An open-source on-chain payments rail
1,746 passing tests · 18 contract pairs live on Base, 17 mirrored across 8+ public testnets · one address everywhere (as of 2026-07-08)
Access0x1: a USD-priced payments router plus capability modules (subscriptions, escrow, refunds, revenue splits, gasless pay-in), each deployed behind an upgradeable proxy at a deterministically mirrored address — the same address on every chain. MIT-licensed, testnet-only by policy until audited. Every test, every deployment, every address is public.
One idea → tested contract, same day
69 unit tests + 2 fuzz invariants · zero handler reverts in 262k fuzz calls
A worked example of the loop at speed: an idea (instant promotional rebates paid inside the settlement transaction, replacing the mail-in rebate and its breakage model) became a production-pattern smart contract with a full adversarial test suite — replay protection proven, funds conservation proven under fuzzing, hostile-receiver paths proven non-blocking — in a single working session. Built and tested, then deployed to Base Sepolia and proven live: real $1 payments settled through the rail on two public testnets, Base and Arc, within the same 24 hours.
The interactive lesson
SEE → VERIFY → LEARN → PROVE, in about a minute
A teaching artifact built for skeptics: read live oracle data from a page with no backend of ours, verify the same record independently on a public explorer, pass a three-question quiz, and optionally write a proof to a testnet — no money, no risk. Built to make the one real idea (a public record no single company owns) tangible.
The method that produced it
The learning isn't the apps — it's the repeatable loop that builds them: ground-truth first, build against frozen interfaces, attack your own output, gate everything, and keep the scoreboard honest.
AI as the workforce, human as the owner
Work is decomposed and fanned out to AI agents — scouts that ground-truth a codebase before anything is written, builders that work from frozen interfaces, reviewers that attack the result. Every agent-reported success is re-verified independently before it counts.
Adversarial verification over trust
Findings and designs survive a refute-by-default pass or they don't ship. The rebate module's row in our own planning catalog was corrected twice by this process — first the naive design (contracts can't listen for events; be the caller), then the cost estimate. The correction trail is kept, not hidden.
Gates, not vibes
Nothing merges without the full gate green (build + tests + formatting + typecheck), and nothing counts as shipped without a canary boot and a health check in the running environment. Deploys roll back automatically on failure.
An honest scoreboard
The operating rule, adopted after learning it the hard way: DONE means deployed and working — verified by a real observed transaction, not by green tests. Our own 100+-app catalog currently scores itself 0 DONE (as of 2026-07-08), in writing, because nothing on it serves real traffic yet. Specs that feel like progress are counted as in-flight.
The on-chain rail runs on public testnets only — no real money — and says so everywhere it appears. Counts on this page are as of the last deploy of this site.