Where I can actually help.
Four ways to work with me, all based on things I’ve shipped in production, repeatedly. Scoped up front, limited in time, and done by me personally.
AI Architecture & Production Review
For founders and CTOs
An independent look at an AI/ML system that is in production or about to be. Does it stay stable under load and budget pressure, are the right primitives in place (ML, LLMs, embeddings, retrieval, agents), and what needs fixing first?
When it makes sense
- The prototype convinced everyone. Now it has to survive real traffic, real budgets, and real failure modes.
- Quality, latency or the inference bill is going the wrong way, and nobody can say exactly why.
- A bigger build, a re-platforming or a key hire is coming up, and you want a second opinion first.
What I look at
- Architecture and data flow, from system diagram to deployed service
- Evaluation: is quality actually measured (offline, A/B tested)?
- Serving, latency and cost per request
- Observability, drift monitoring, failure modes
- Reproducibility, CI/CD, test coverage that means something
- Build vs. buy, and whether the team can run it
What you get
A written report with prioritised findings and a session with your team to go through it. Concrete enough that your engineers can start on the first item the next day.
Format
Usually two to three weeks, part-time. I read code, docs and dashboards, and talk to the people who build and run the system.
AI Technical Due Diligence
For VCs and investors
A technical assessment of a company whose value depends on its AI. Does the technology match the pitch, and what would it take to scale it?
When it makes sense
- The AI is the product, and you need to know how much of it is real.
- Own technology, or a thin layer over someone else’s model? You want to know before you sign.
- A portfolio company’s AI roadmap keeps slipping, and you want an independent view on why.
What I look at
- What’s behind the claims: models, data, evaluation
- Defensibility: what’s hard to copy and what isn’t
- Scalability and what inference costs per unit
- Engineering practice and depth of the team
- Dependencies and risk: model providers, OSS licences, security
What you get
A short memo an investment committee can read: findings, red flags, open questions for the founders and a clear overall call. With realistic feasibility calls, and pushback where the pitch over-promises. The technical detail goes in the appendix.
Format
Under NDA, timeline agreed up front. Based on documents, code access where possible, and conversations with the technical team.
Technical Advisory
For founding engineers and CTOs
A sparring partner for a few hours a month. My philosophy: fast prototypes for validation, robust system engineering for production. Use the right tool, not the existing.
Typical topics
- Architecture decisions
- Build vs. buy
- Roadmap
- Design reviews
- Interviewing the ML engineers you want to hire
Agentic Engineering Consulting
For engineering teams
Practical coaching and architecture for teams adopting agentic coding tools — Claude Code, Cursor, Copilot, custom agents. When implementation gets faster, everything piles up before and after it: in decisions, reviews and approvals. So most of the work is around the agent. Same engineering rigour applied to a brand-new toolchain, so you don’t end up with code nobody on the team really understands anymore.
Around the agent
- CI integration and automated PR review
- Test generation and QA
- Refactor agents
- Guarded production changes: who signs off on what
- Observability that closes the verification loop. An agent that can read logs is worlds better than one that can’t.
Not my thing: full-time contracting, or bringing a whole team to build the product. If that’s what you need, I’ll tell you in the first call, and I usually know who to ask.
