Custom software — from an internal tool to a multi-tenant platform

Software projects usually fail at the seams — between the designer and the developer, between the frontend and the backend, between the people who built it and the people who have to run it. The way around that is not better handoffs. It is fewer of them.

Everything here is built end to end by one engineer: data layer, API, interface, deployment, and the operations afterward. There is no integration phase because there is nothing to integrate.

What gets built

• Internal tools that replace a spreadsheet nobody trusts anymore
• Customer portals and account systems on top of existing business software
• Multi-tenant SaaS platforms with isolated data per customer
• Integration layers and sync bridges between systems that do not talk
• Scheduled jobs and background workers that keep data current without anyone clicking anything

Proof, not promises

The platform this site runs on is the reference implementation: one process serving more than a dozen isolated brands, each with its own database, theme, domain, and feature set. Alongside it run a self-hosted shop management system replacing incumbent SaaS, a developer API with an open-source client published to npm and the official MCP registry, and a broadcast pipeline that writes, voices, and streams a channel around the clock. The public catalogue is on the work page.

How AI actually gets used

AI features are built where they create leverage and left out where they do not. The standing rule on every product here: intelligence runs on a schedule in the background and caches its results, so the interface is instant and there is never a "Generate" button asking a user to wait on a model. Spend is metered and capped rather than open-ended.

Operating it afterward

Software that nobody operates degrades. Builds here come with the deployment, monitoring, backup, and update path already in place, because the same person is going to be the one woken up if it breaks.