AI

AI features and tooling that work in production, securely.

Talk to an engineer
Problems we solve
“We want AI in the product but don’t know where to start.”
“Engineers use coding agents with no guardrails.”
“GPU costs are out of control.”

What we do

Advise

AI consultancy

An assessment of your AI setup, with findings ranked by risk and effort.

Build

AI delivery

We design and build it with your team, in your accounts, documented as we go.

Run

AI support

Monitoring, patching and upgrades, with incident response agreed per contract.

Teach

AI training

Workshops and pairing on AI, built around your own systems.

Four kinds of AI work

AI in your productLLM features, search over your own data, agents and integrations built into your apps.
AI-assisted engineeringAdopting coding agents safely across a team: workflows, guardrails, review practice and training.
AI infrastructure / MLOpsGPU and cloud setup, model serving, pipelines and cost control for AI workloads.
AI strategy & governanceFinding the right use cases, and the data privacy, security and policy to go with them.

Tools we work with

Models & APIs

  • Anthropic Claude
  • OpenAI
  • Amazon Bedrock
  • Azure OpenAI
  • Vertex AI

Retrieval & data

  • pgvector
  • OpenSearch
  • Pinecone
  • Qdrant
  • LlamaIndex

Agents & integration

  • Model Context Protocol
  • LangGraph
  • tool calling

Serving & infrastructure

  • vLLM
  • Ollama
  • NVIDIA GPUs
  • KServe
  • Ray
  • SageMaker

Evaluation & monitoring

  • Langfuse
  • promptfoo
  • OpenTelemetry

AI-assisted engineering

  • Claude Code
  • GitHub Copilot
  • Cursor

How an engagement runs

  1. 1 A call. 30 minutes with an engineer to understand the problem.
  2. 2 A look. Read-only access or a walkthrough, so we scope from facts.
  3. 3 A proposal. Scope, deliverables and a price model: fixed scope, day rate or retainer.
  4. 4 The work. In your accounts and repositories, alongside your team.
  5. 5 A handover. Documentation and a session with your team, so you’re not dependent on us.

Proof

A recent AI engagement. We keep client names private, but the problem, the work and the outcome are exactly as they happened.

See our work
Case study · anonymised
UK B2B software company
Challenge
We want AI in the product but don’t know where to start. That was how the client put it on the first call.
Approach
We ran a short discovery to pick one use case, built search over the client’s own documents with access controls, and shipped it behind a feature flag.
Stack
LLM API · retrieval (RAG) · PostgreSQL · access controls
Result
One AI feature in production, with data-handling rules the client’s security lead signed off.

Questions

Do you work in our accounts or yours?

Yours. Everything we build lives in your cloud accounts and repositories from day one, so there is nothing to migrate when we finish.

Can you work alongside our own team?

Yes, and we prefer it. We can run a project ourselves, embed engineers in your team, or pair with your people so they pick it up as we go.

How is this priced?

Depending on the work: a fixed-scope price, a day rate or a monthly retainer. We agree it in a written proposal before anything starts.

Tell us what’s broken, slow or expensive.