Indie Products
Five indie products, five implementation write-ups
These are the products where I got to make the calls that client work rarely allows — and carried them through on my own responsibility. Three of the five haven't shipped yet, so what's written here is not user counts or revenue, only numbers you can count from the code. If it can't be counted, it isn't here.
- Web app live
Designing an AI that never states a fact — a deterministic engine, with the LLM only writing prose
Palm features extracted by a VLM, the chart computed deterministically, and the LLM allowed to do one thing: turn those facts into sentences — whose citations are then machine-checked. How Palmia is built.
Read the write-up - Pre-orders open
One algorithm, implemented twice — in TypeScript and embedded C — and stitched together by CI
An IoT early-warning system that alerts on the rate of change in water temperature and TDS, with the same decision logic in the cloud and on the device, pinned by golden-fixture contract tests. How Uomamori is built.
Read the write-up - Live
How to build a site AI answer engines will cite — 371 articles across 7 languages
An llms.txt, a CC BY 4.0 machine-readable dataset, and a test that checks the *deployed* robots.txt. How ProxyFacts was designed for answer engines first.
Read the write-up - Coming soon
A push pipeline written to assume failure — in nothing but SQL
No external worker infrastructure: the dead-letter queue, the retry curve and the reaping all live in Postgres. How memofu's delivery layer and privacy design are built.
Read the write-up - Coming soon
The bug I shipped is still in the code
Making an offline review queue idempotent, keeping an async pipeline under API Gateway's 30 seconds, and encoding a patent freedom-to-operate analysis as constraints in code. How MemoryHack AI is built.
Read the write-up