Service
Retrieval, summarisation and assistants wired into a product that already works — with a cost ceiling and a plan for being wrong.
We start every AI conversation with two questions: what does a wrong answer cost, and what does a right answer save? If the first number is high and the second is small, we will recommend not shipping a model at all — a rules engine, a better search index or a well-designed form is often cheaper, faster and more honest.
When a model is the right answer, we treat it like any other unreliable dependency. There is an evaluation set before rollout, a cost ceiling per request, caching for repeated work, a timeout, and a defined behaviour for the moment the provider has an outage. Output that affects money or health goes through a human review path.
We also write down what the feature costs per active user per month, because AI features have a habit of quietly becoming the largest line on the infrastructure bill.
Describe the outcome you want instead of the deliverable you think you need. We will tell you which service fits — or tell you honestly that you do not need us yet.
The whole business in software form: onboarding, accounts, permissions, subscriptions, admin tooling and the dashboards you need to run it.
Native Swift and SwiftUI work with App Store review treated as part of the build, not an afterthought.
Kotlin and Jetpack Compose, built for the device spread that actually shows up in your analytics.
Let’s scope it properly
Send us the shape of your idea and we will come back with a written scope, a realistic timeline, a fixed price band and the risks we would want to kill first. No decks, no discovery invoice, no pressure.