Assistants and conversation
Assistants that hold context, use your data through retrieval, and call real tools rather than guessing. Built with explicit limits on what they can see and do.
Services
We work as a small senior team on a handful of projects at a time. That means you get the people who scoped the work actually building it.
Application development
Web applications and native mobile apps, built with React, Next.js, TypeScript and a Postgres‑shaped data layer. We ship to the App Store, Google Play and the web, and we set up the deployment pipeline so releases are routine rather than an event.
Everything is written to be handed over. Documented architecture, readable code, infrastructure as configuration and accounts in your name — so you are never locked into us by accident.
Discovery and scope 1–2 weeks
Prototype 2–3 weeks
Build cycles 2 weeks each
Launch and run Ongoing
AI engineering
We use current frontier models where they genuinely improve the product, and say so plainly when they don’t.
Generation with structure
Generated documents, lessons, summaries and reports that stay coherent across a whole body of work — the hard part is consistency, not a single good output.
Matching and classification
Scoring, ranking and routing, with the reasoning surfaced so a user can see why they got a result.
Evaluation harnesses
Test sets and scoring that run in CI, so a prompt or model change can’t quietly degrade quality.
Guardrails and fallbacks
Input and output filtering, rate limits, cost ceilings and a defined behaviour for when a model is unavailable.
Governance and assurance
Available inside a build, or on its own if you have a product already live and a deadline approaching.
Data protection
Records of processing, lawful basis, DPIAs, retention schedules and subject access handling.
AI documentation
Model cards, intended use, known limitations, human oversight design and an audit trail of changes.
Accessibility
WCAG 2.2 AA audit with a prioritised remediation plan and retesting after the fixes land.
Security review
Threat modelling, dependency and secrets auditing, authentication and authorisation review, and coordination of third‑party penetration testing where the risk profile calls for it.
Policy set for launch
The public documents a product needs on day one — privacy, terms, cookies, acceptable use, AI transparency — written to match what the software actually does, then reviewed by your legal advisers.
How we charge
Three ways of working. We’ll recommend the one that fits the certainty of your scope, not the one that bills most.
Discovery sprint
Fixed price, one to two weeks. Scope, prototype, technical approach and a costed plan. You keep everything whether or not you continue with us.
Fixed‑scope build
A defined product for a defined price, staged across milestones. Best once discovery has removed the big unknowns.
Ongoing team
A monthly retainer for continuous development, support and improvement after launch. Cancellable with notice.
Not sure which you need?
Send a paragraph about the problem. We’ll reply with what we’d do first and a rough range, at no cost.