Services

Senior people, narrow scope, visible progress.

We do four kinds of work. Each one is led by someone who has shipped that exact thing before, and each one ends with you owning the result outright.

Applied AI

The gap between an impressive prototype and a system your customers depend on is mostly evaluation, retrieval quality and failure handling. That gap is where we live.

Talk about an AI build

What we build

  • Retrieval systems over messy internal corpora — chunking, hybrid search, reranking, citation you can audit.
  • Agentic workflows with real tool boundaries, retries and human checkpoints where the stakes require them.
  • Evaluation harnesses first: a regression suite for model behaviour, so a prompt change can't silently break production.
  • Cost and latency budgets tracked per request, with caching and model routing that keeps them honest.
  • Fine-tuning and distillation when — and only when — the numbers justify it over a better prompt.

What we won't do

Bolt a chat box onto a product that doesn't need one. If a lookup table and a well-designed form solve it, we'll tell you that and charge you for a much smaller project.

Product engineering

End-to-end delivery for teams that need the whole thing built, not a piece of it staffed.

Typical shape

  • Data model and API design before a line of interface code.
  • Typed from database to browser; the compiler catches what tests would otherwise have to.
  • Design and build in the same room — no throwing mockups over a wall.
  • CI, preview environments and one-command deploys on day one, not month three.
  • Accessibility and performance treated as acceptance criteria, not a later ticket.

Stack we reach for

TypeScript, React, Node, Python, Postgres, and whichever cloud you're already committed to. We're not religious about it — we're religious about you being able to hire for it afterwards.

Platform & scale

For when the product works, the users arrived, and now the bill and the p99 are both going the wrong way.

Where we usually find it

  • Query patterns that were fine at ten thousand rows and are not fine at ten million.
  • Missing observability — you can't fix what you can't see, so we instrument before we optimise.
  • Cloud spend growing faster than usage, almost always in egress, idle capacity or over-provisioned managed services.
  • Deploys nobody trusts, so nobody deploys, so every release is enormous and terrifying.

How we deliver it

Measured before-and-after on every change. If we can't show you the graph moving, we didn't do the work.

Technical due diligence

For investors and acquirers who need to know what they're buying before the wire goes out.

Two weeks, written down

  • Codebase health, architecture risk and the real cost of the next twelve months of roadmap.
  • Team assessment: key-person risk, hiring reality, what happens if the founding engineer leaves.
  • Security and compliance posture at a level a board can read.
  • A blunt list of what we'd fix first, in order, with rough effort attached.

Delivered as a written report plus a live session. We don't soften findings to keep a deal warm.

Engagements

Three ways to work with us.

Indicative ranges. Real numbers come out of discovery, and we'd rather quote late and accurately than early and wrong.

01 — Discovery

Two weeks, fixed

Architecture, risk register, a spike of the hardest part and a costed build plan. Yours to keep and take anywhere.

Fixed price2 weeks

02 — Build

Delivery sprint

A dedicated pair or trio, weekly deploys, a shared board and direct access to the engineers. Typically eight to twenty weeks.

Monthly8–20 weeks

03 — Embedded

Inside your team

Our engineers on your standups, your repo, your on-call rotation, for a quarter or two. Knowledge stays when we leave.

QuarterlyRetained

Tell us what you're building.

A paragraph is enough to start. We reply within one business day.