01 / In-house products
Products we built for ourselves, running live.
We build our own products for the same reason we tell founders to ship early: it keeps our judgement honest. Everything here is live, maintained by us, and paid for by us.
02 / The shelf
Four in-house products, each with real users on it.
Product name
One line on what it does and who it is for. Replace this with the real description when you are ready.
Since 2024Self-funded
Read the breakdown →Product name
One line on what it does and who it is for. Replace this with the real description when you are ready.
Since 2025Paying users
Read the breakdown →Product name
One line on what it does and who it is for. Replace this with the real description when you are ready.
Since 2025Used on every build
Read the breakdown →Product name
One line on what it does and who it is for. Replace this with the real description when you are ready.
Since 2026In growth
Read the breakdown →03 / Library
What we learned building them, written down.
ArchitecturePostgres held our first 100,000 documents just fineWe reached for a vector database on instinct, measured it, and stayed on Postgres for another year. What the numbers actually said.Aug 2026 · 7 minAI engineeringCache the retrieval, not the answerCaching model output looks like the cheap win. In practice the retrieval step was the expensive one, and it was the one that repeated.Jul 2026 · 6 minReleaseThe three numbers we check before every releaseNot test coverage. Error rate on the main workflow, p95 on the slowest query, and how long a rollback takes.Jun 2026 · 5 minEngagementWhat we actually do in week oneNo discovery theatre. A running deployment, a data model on paper, and a written list of what we are not building.May 2026 · 8 min
04 / Start here