I help teams test and evaluate AI features before they ship — chatbots, assistants, classification models, anything with an LLM behind it — so you catch the awkward, risky or plain wrong answers before your users do.
Every engagement starts with a short call — no fixed packages, scoped to what you're actually shipping.
No price list up front — projects vary too much for that. What you do get is a clear process, start to finish.
Tell me what you're building and what's worrying you about it. No cost, no commitment.
Based on that call, I scope the work and send a clear quote — what's tested, what you get, and when.
The hands-on work, detailed below — testing your feature and writing it up so your team can act on it.
You get the findings, and I'm around for follow-up questions once you've had time to read them.
Before anything gets written up, it needs to be properly tested. This is the hands-on part — working through your AI feature the way a real user (or someone trying to break it) would.
Does the AI actually do what it's supposed to, across realistic prompts and the inputs your users will genuinely send?
Same question, different day — does the answer still hold up, or does it quietly drift?
How does the model behave across different inputs, names, and contexts? Where does it treat people inconsistently?
Prompt injection, jailbreak attempts, data leakage — the things a user shouldn't be able to pull off, but might try.
The odd, unexpected inputs nobody thought to test — usually exactly where the real problems hide.
Testing without clear reporting is just testing. Everything above gets written up so your team — technical or not — knows exactly what to do with it.
Each issue written so a non-technical stakeholder understands the risk, not just the technical detail.
What was tested, what passed, what didn't, and what to prioritise first.
A clear plan for what gets tested and why, scoped to your feature before testing even starts.
Where it's relevant, findings mapped against frameworks like the UK FCA's AI principles.
Still not fixed prices — scope and timeline decide that — but a sense of what each kind of engagement looks like.
A thorough pass before you ship. One feature, one findings report, one clear answer on whether it's ready.
Good fit if: you have something close to launch and need a second opinion before it goes out.
Testing alongside your iterations as a feature gets built out, through to launch — not bolted on at the end.
Good fit if: you're actively building and want testing built into the process as you go.
Regular testing and reporting on a standing basis — working like part of your team, not a one-off hire.
Good fit if: you ship AI features often and want testing as a standing part of how you work.
Every project is scoped individually — feature size, risk level and timeline all matter. After the first call, you get a clear quote before any work starts. No surprise invoices.
A one-off evaluation usually runs 1–2 weeks. Launch support and ongoing work are scoped to your own release schedule, not a fixed clock.
Yes — fully remote, based in London and working UK hours. Most of this happens over a call and shared documents.
Chatbots, assistants, classification models — anything with an LLM or ML model behind it that your users interact with directly.
A few quick details help me reply with something useful, not a generic "tell me more."