Engagements,
not a menu.
Every engagement works on the same six-layer system. They differ in where you start: checking something already live, proving a new idea, building it for production, putting one platform under every team, or keeping a live system improving.
Embedded Engineering
Engineers who keep a live AI system improving: better quality, lower cost and new capabilities, each change measured.
- STARTING POINT
- A live system
- ENDS WITH
- A system that keeps improving
- BEST FOR
- Teams that need senior AI engineering continuously, without hiring a full team
- LEADS INTO
- Continues while it is useful
- 01Regular evaluation reportDOC
- 02Model and prompt upgradesCODE
- 03New capabilities, scopedSOFTWARE
- 04Cost and reliability reviewREADOUT
3 phases, each with a checkpoint.
- WE DO
- Set up evaluation and cost baselines
- YOU PROVIDE
- The metrics that matter to the business
- CHECKPOINT
- Metrics agreed
Five ways in, one system.
| Compare | |||||
|---|---|---|---|---|---|
| STARTING POINT | AI in production | An idea | A prototype | AI across many teams | A live system |
| YOU ARE HERE IF | AI is live, but it’s fragile, slow or expensive | You have an idea, not a system | The prototype worked; production didn’t | Several teams use AI; cost, risk and duplication are growing | It’s live and needs to keep getting better |
| LAYERS BUILT | None. Review only | L2 L3 L4 L5 | L0 L1 L2 L3 L4 L5 | L0 L1 L2 L3 L4 | L1 L2 L3 L4 |
| YOU LEAVE WITH | A prioritised plan | A tested prototype | A system you own | An AI platform you own | A system that keeps improving |
| LEADS INTO | Production Build or Embedded Engineering | Production Build | Embedded Engineering or AI Platform | Embedded Engineering | Continues while it is useful |
What we don’t sell.
- Chatbot widgets bolted onto a website
- AI strategy decks with no system behind them
- Engineers rented out by the hour
- White-labelled resale of someone else’s product
Asked before signing.
Yes. The audit ends with a written plan you own. Many teams fix things themselves from it; some ask us to help with the parts that need deeper engineering. There is no obligation either way.
You do. Everything we build lives in your repositories and runs in your infrastructure: code, prompts, evaluation sets, configuration and logs. Models are rented from providers; the intelligence around them is yours.
We read your brief, have a short call with the people who know the workflow, and send a written proposal with the scope, deliverables and checkpoints before anything starts. No work begins on an open-ended estimate.
A named owner who will run the system after we leave, access to the relevant systems and data, and regular time from someone who knows the workflow well. We work with your engineers, not around them.
Every proof has a quality bar agreed at the start. If the prototype does not meet it, you get the evaluation results and a clear explanation of why, before you have spent anything on production. That is a useful result, not a wasted one.
Whichever fit the task and your constraints. We build model-agnostic systems that can route between providers or run on private models, deployed in your cloud account, so switching later is a configuration change.