Most AI projects start with more unknowns than an honest quote can cover. So I don’t price the whole project up front. Each step has a fixed scope and a fixed price, and the next step is priced once the previous one has shown what it needs to be.
How it works
Four steps. You decide at the end of each one whether to take the next.
- 0 Free review Free You send the problem. I reply with a short written read, so our first conversation starts from something concrete.
- 1 A first package Fixed price One of the starting points below. Small, fixed in scope, usually one to two weeks.
- 2 Phased work Priced per phase Design or build, one phase at a time. You get a range for the whole project and a fixed price for the next phase only.
- 3 Ongoing Monthly Optional. Advice or support after handover, for as long as it is useful.
Starting points
Choose by where you are, not by which service you think you need. If you can’t tell, the free review will.
Exploring an idea
Readiness Assessment
You are considering an AI initiative and want to know whether it is worth the budget before you commit.
You get
- A written recommendation (go, change course, or stop) with the reasoning behind it.
- An honest look at the data, the process around it, and who would own the system.
- What happens when the system is wrong, and whether you can live with that.
- If it is a go: the scope and price of the next phase.
Done before: readiness assessments and success criteria for prospective AI contracts in wildlife monitoring, smart agriculture and smart cities.
- Typical length
- 1–2 weeks
- Price
- Quoted after the free review
Reviewing a proposal
Proposal Review
A vendor or contractor has sent you a proposal for an AI system, and you want an independent read before you sign.
You get
- Risks, gaps and unrealistic assumptions, marked against the proposal itself.
- Acceptance criteria and a definition of “done” you can hold them to.
- The questions to ask them before you commit.
- Typical length
- 3 working days
- Price
- Quoted after the free review
Ready to build
Design Sprint
You have decided to build and need a design that a team, yours or mine, can build from.
You get
- Architecture, component boundaries, data requirements and failure handling, including where not to use a model.
- Success criteria and an evaluation plan, agreed before the build starts.
- A build sequence with a cost range, and a fixed price for the first build phase.
- A small prototype of the most uncertain part, when one is needed.
Done before: the Persian voice assistant, with its edge–cloud split, four services, and a 40 KB per-user model.
- Typical length
- 2–3 weeks
- Price
- Quoted after the free review
Already built, and struggling
LLM Cost & Latency Audit
Your LLM feature works, but it is too slow or too expensive to run.
You get
- A trace of where the time and tokens actually go, step by step.
- The calls that can be removed, cached, or handed to a smaller model or a plain lookup.
- Before-and-after measurements on a sample of your real traffic.
Done before: removing one model call cut median latency from 161 s to 36 s and doubled the correct answers.
- Typical length
- 1 week
- Price
- Quoted after the free review
RAG Health Check
Your retrieval-based assistant gives answers that are wrong, vague or inconsistent.
You get
- Failure analysis on about 30 of your real questions, each wrong answer traced to its cause.
- Ranked fixes, with the expected effect of each.
- A set of test questions you keep, so future changes are measured rather than guessed.
Done before: traced every wrong answer of a RAG chatbot to one of four causes, then measured each fix separately.
- Typical length
- 1 week
- Price
- Quoted after the free review
After the first step
Larger work is never priced as one block.
Design and build phases
You get a cost range for the whole project and a fixed price for the next phase. The range narrows as each phase answers questions the previous one couldn’t.
- Unit
- One phase, or a two-week sprint
- Price
- Quoted per phase
Advisory retainer
A set number of hours each month for architecture questions, reviews and decisions, during a build or after handover. Cancel with one month’s notice.
- Unit
- Monthly, paid in advance
- Price
- Quoted on request
How payment works
- Fixed prices, not hours. I track my time to keep my estimates honest, but you pay for the agreed scope. If you need hourly billing, I offer it with a cap and weekly timesheets.
- Each phase is paid before it starts. For phases longer than two weeks, half at the start and half on delivery.
- The first package carries a guarantee. Half up front, half on delivery. If it isn’t useful to you, you don’t pay the second half.
- New scope is never added silently. Anything new during a phase is written down and either priced as a change or moved to the next phase.
- Either of us can stop at the end of any phase. You keep everything you have paid for.
- Currency and payment method are agreed in writing before any work begins.
What we agree on, in writing
No long contract. Before each phase you get a one-page proposal, and a reply saying “agreed” is enough. It always covers:
- The problem and the scope
- What is out of scope
- Deliverables, and how we both know each one is done
- Timeline
- Price and payment schedule
- How changes are handled
- Who owns the code and outputs
- Confidentiality
- How either side can stop
If your organisation needs its own contract, that’s fine too; I’ll read it alongside the proposal.
Not sure which fits? The review is free and will tell you.