Four things, done properly.
React Native for both stores, with the native modules written by hand when the library does not exist.
Next.js applications with real authorisation models, not a role column and hope.
Node and Python services, queues that survive restarts, and schemas designed for the queries you actually run.
Terraform, Kubernetes, and monitoring that pages a human only when a human is needed.
Four steps to a signed scope.
Share context
A paragraph, a repo, or a call. Whatever you have.
Talk to the engineer
Thirty minutes with the person who would do the work. No account manager.
Get a real plan
Written scope, sequencing, risks and what we would not build. Yours either way.
Start small
A fixed-price two-week piece so you can judge us on output rather than a pitch.
Amour acknowledges a message before moderation runs.
A spinner on every send makes a chat app feel broken. The send path writes the message as pending and returns; moderation runs behind it off the queue.
A message can appear and then be pulled. We had to build retraction into the client, the read model and the notification path — none of which a synchronous design would have needed.
Two kinds of client.
White-label, under your name.
You keep the client relationship. We stay invisible: your repos, your Slack, your standups if you want us there. Commits land under whatever identity you set up, and we never contact your client directly unless you ask us to.
- Fixed-price scoping so you can quote with a margin
- We write the handover docs your client will read
- NDA and non-solicit signed before any specifics
Embedded, with backend depth.
Most product teams can hire front-end. The gap is the person who has run a queue under load, argued with a Postgres query plan, and been paged at 3am for their own cluster. That is the seat we fill.
- Backend, data and infrastructure, not staff augmentation
- Four hours of live overlap, async the rest of the day
- Every architecture decision comes with its stated cost
Two projects, written up in full.
Our own dating app on Android. gRPC services over Postgres and Neo4j, moderation that runs off the queue instead of in front of the user, and a K3s cluster we operate ourselves.
A facilities operations platform for a healthcare estate: a web app, a mobile app for supervisors, a phone page for technicians, and one backend under all three. Hourly work-order sync into ClickHouse, repeat-fault detection, and a weekly PM pulse written by Claude.
From the team that runs on it.
We went from exporting spreadsheets by hand to a platform our whole facilities team runs on. Supervisors start the day with their overdue list, the Monday huddle opens with the pulse already written, and if a dashboard falls behind, we hear about it before anyone acts on stale numbers.
Two we can show. The rest are under NDA, and a logo we cannot point to real work behind is worth nothing to you.
Infrastructure we operate in production today, not partner badges. Every one of these appears in a case study on this site.