Consulting
Since February 2026 I have run an independent practice, taking on focused engagements in AI evaluation, model interpretation and the engineering around them — alongside my research.
Areas of work
Each engagement is scoped to a concrete question and ends in evidence the team can inspect.
Model evaluation
Define evaluation tasks, compare models or configurations under controlled conditions, and locate where performance breaks down.
Model interpretation
Examine what drives a prediction, test the assumptions behind an explanation, and separate useful evidence from a convincing story.
Agentic systems
Build agent workflows with bounded tasks, independent review and recorded trajectories, then measure reliability, latency and cost against agreed tests.
Technical AI governance
Turn agreed policies into technical controls and documented evaluation evidence that a team can inspect and maintain.
How engagements run
Established professional practice, written down before work starts.
- Scoping
A short call to understand the problem, constraints and what a useful result would look like.
- Written proposal
Scope, deliverables, acceptance criteria, timeline and fees agreed in writing before work starts.
- Confidentiality
A non-disclosure agreement where needed. Client names, data and commercial details stay private by default.
- Milestones
Work delivered in reviewable stages, each with evidence the client can inspect — code, results and their limits.
- Handover
Documentation, reproducible code and a final report, so the work stays usable after the engagement ends.
Client work and confidentiality
Some of the projects on this site were carried out for clients. Client names, data and commercial terms are confidential, so they are not listed here. Where an agreement allows, more detail can be discussed directly.
Get in touch
For an engagement, a role or a research collaboration.