Consulting
AI engineering, delivered by the people who will do the work.
We take on a small number of engagements a year so each one gets a founder's attention. Here is what we do, how an engagement runs, and who it suits.
LLM application development
From a working prototype to a service your team runs: prompts, tools, guardrails, observability and the boring parts that make it dependable.
Agent systems
Multi-step agents that use tools, keep state and know when to stop and ask. We have shipped these for our own products and know where they break.
Retrieval and evaluation
Retrieval pipelines that are measured rather than assumed, with evaluation suites so you know when a change made things worse.
Model deployment on Google Cloud
Cloud Run, Vertex AI, Firestore and Postgres, with cost controls from day one. We run our own products on the same stack.
Prototyping and technical due diligence
A week to find out whether the idea works before you commit a quarter to it, or an independent read on someone else's AI claims.
Engagement models
Fixed-scope project
A written scope, a fixed price and a delivery date. Best when you know what you want built.
Monthly retainer
A set number of days a month for ongoing work, with the same engineer throughout.
Advisory hours
Architecture reviews, code reviews or a second opinion, billed by the hour with no minimum.
How it runs
- 1
A thirty-minute call to understand the problem. No charge, no slides.
- 2
A written proposal within a week: scope, approach, price, and what we will not do.
- 3
We build in the open, in your repository if you prefer, with a working demo every week.
- 4
Handoff with tests, documentation and a recorded walkthrough, so the work survives us leaving.
Fit
A good fit
- You have a specific problem and want it solved well, not a slide deck about AI.
- You are on Google Cloud, or open to it.
- You want the code, the tests and the documentation, not a dependency on us.
Probably not
- You need a large team on site next week.
- You want a model trained from scratch.
- The brief is to make something look AI-powered rather than work better.
Ready when you are.
Request a quote