About
I build and evaluate AI systems for settings where being wrong matters
I’m Nick Lamb, a biochemistry PhD who spent more than 20 years in medical communications before moving into building AI full time in 2022. Today I ship AI into regulated healthcare, and design the experiments and open-source tools that check whether AI answers hold up against evidence.
Open to full-time applied AI & evaluation roles · London / Oxford / UK-remote
The path
From the lab bench to shipping AI
- The benchPhDPhD, Clinical Biochemistry · Imperial College London
Redox signalling in patients with acute lung injury. Three years at the bench, including GC–MS.
- The domain20+ yearsMedical & scientific communications
Writing accurate, engaging content for both HCPs and patients, across medical affairs and promotional work.
- The codeFull timePharmaTools.AI
Building AI full time: products, open-source evaluation and privacy tooling, and research on autonomous agents.
- The futureActiveA full-time applied AI or evaluation role
Evaluation, forward-deployed or applied AI engineering, in a team that takes evidence seriously.
What I do
Four things, each with something you can check
OpenGATE
An open-source framework that tests whether an AI system can prove its answers from its sources, with gold sets, scorers and CI gates. In a 61-claim benchmark my LitRAG pipeline caught every hallucination; RAGAS and DeepEval missed some.
See OpenGATE → Agent researchObserver Zero & preregistered studies
235 seeded, manifest-frozen runs on how LLM agents gather evidence and revise beliefs (combined paper in peer review), plus a preregistered study of acceptance gates for an LLM that rewrites its own code.
See Observer Zero → Shipping in regulated healthcarePatiently AI
An MHRA-registered Class I medical device that explains medical notes in plain language, used in 100+ countries and validated in a peer-reviewed study in Frontiers in Digital Health (2026).
See Patiently AI → Privacy & agent toolingRedacta & PubCrawl
Redacta pseudonymises patient data before AI sees it; PubCrawl gives AI assistants real PubMed access. Both are listed in Anthropic’s official MCP directory.
See the MCP servers →How I work
Principles I hold the work to
- Check answers against evidence, not plausibility
An output counts as right when it can be traced to a source, not when it sounds right.
- Freeze the design before the data
Confirmatory studies are preregistered and frozen at a public tag before the first run.
- Publish what didn’t work
Null results and gamed metrics go in the write-up, because they are often the most useful finding.
- AI-assisted, human-owned
I use Claude Code daily for implementation, with every change checked against tests and evaluation gates. Experiment design and interpretation stay with me.
Credentials
Education
- PhD, BiochemistryImperial College London
- BSc (Hons) Applied Biology, First ClassUniversity of Greenwich
Certification
- CMPPCertified Medical Publication Professional
Awards
- Communiqué Award 2025Progress Award, Healthcare & Scientific Communications
- PMEA 2025 ×2, HTN and Best Mobile App Awardsfor Patiently AI
- Open Targets Hackathon 2025Audience Choice (BiomarkerFinder)
Supported by
- Claude for Startups, Google for Startups, AWS, Microsoft for Startups, NVIDIA Inception, ElevenLabs and Perplexity startup programmes
Advisory
Alongside my search for a full-time role, I take on a small amount of advisory work where I can add something specific: patient-facing AI (clinical safety, regulatory route, privacy and plain-language design, from registering Patiently AI myself) and evaluation for AI that has to prove its answers.
Hiring for applied AI or evaluation?
I’d be glad to talk. London, Oxford or UK-remote.