Profile
Nick Lamb, PhD, CMPP
Applied AI & evaluation for healthcare
Summary
Biochemistry PhD with more than 20 years in medical and scientific communications, building and evaluating AI full time since 2022. I ship AI into regulated healthcare, including an MHRA-registered Class I medical device, and build the open-source tools and preregistered experiments that check whether AI answers hold up against evidence.
Looking for
Full-time, permanent: Applied AI Engineer, AI Evaluation, Forward Deployed Engineer, AI Product Builder
London (hybrid or onsite), Oxford or UK-remote. Available at short notice.
What I bring
- Evaluation that holds upGold sets, deterministic scorers, CI regression gates and adversarial benchmarks, with no LLM judge where ground truth exists.
- Shipping in regulated healthcareMHRA Class I registration, peer-reviewed clinical validation, and privacy built into the architecture.
- Rigorous agent researchSeeded, manifest-frozen and preregistered experiments on how LLM agents reason, revise beliefs and game metrics.
- Scientific and clinical fluencyTwo decades turning clinical study reports and literature into accurate material for HCPs, patients and pharma teams.
Selected work
OpenGATE
Evaluation & verification · open sourceA verification framework for evidence-grounded AI: gold sets, deterministic scorers and CI gates. Four production systems run on it in CI, where it has caught silent parse failures, a citation-parser inconsistency and a simplifier dropping an antibiotic dose. In a 61-claim benchmark my LitRAG pipeline caught every hallucination; RAGAS and DeepEval missed some. Ships on npm and PyPI, as an MCP server and as a GitHub Action.
Observer Zero & RSI Loop
Agent research · preregistered235 seeded, manifest-frozen runs on how LLM agents gather evidence and revise beliefs, in a world where only the simulator knows the ground truth; combined paper in peer review. A companion preregistered study compared four acceptance gates for an LLM that rewrites its own code over 120 trajectories: the visible-score gate was gamed on 10 of 10 seeds, and only an absolute behavioural envelope improved hidden truth.
Patiently AI
MHRA Class I medical deviceExplains clinical letters, results and discharge summaries in plain language, in 18 languages on web, iOS and Android, used in 100+ countries. Peer-reviewed validation (Frontiers in Digital Health, 2026): reading level fell by 2.96 grade levels, 87.3% of expert safety assessments judged outputs clinically safe, and patients preferred the AI explanations in 70% of comparisons. Five national awards. Moved to a new underlying model using pre-specified bridging criteria and blinded review.
Redacta & PubCrawl
Privacy & agent tooling · Anthropic MCP DirectoryRedacta is a privacy boundary for AI agents: it swaps patient identifiers for labelled tokens, keeps the map that reverses them outside the model’s context, and checks output on the way back. Adversarially benchmarked, with an MCP server, a Kubernetes deployment for self-hosting and an iPhone app. PubCrawl gives AI assistants real PubMed, Europe PMC, trial and drug-label access. Both are listed in Anthropic’s MCP Directory.
Selected publications
- Validation of an AI-powered mobile application for personalizing medical note explanations: a mixed-methods evaluation · Frontiers in Digital Health, 2026 · journal
- Translation, not Interpretation: Rethinking Language Model Design for Healthcare · SN Comprehensive Clinical Medicine, 2026 · journal
- Observer Zero: Do LLM Agents Form Epistemic Communities? · preprint, in peer review
- Evaluation as Infrastructure: Deterministic Regression Gating for Evidence-Grounded AI Systems in Production · preprint
- A Preregistered Comparison of Acceptance-Gate Architectures for an LLM That Rewrites Its Own Code · preprint
Background
Education & certification
- PhD, Biochemistry · Imperial College London (three years at the bench)
- BSc (Hons) Applied Biology, First Class · University of Greenwich
- CMPP · Certified Medical Publication Professional
Medical & scientific communications
20+ years writing accurate, engaging content for HCPs and patients across medical affairs and promotional work, regularly built from clinical study reports. Clients and agencies include AstraZeneca, Novartis, Astellas, Havas, Ogilvy and Parexel.
Awards
- Communiqué Award 2025 · Progress Award, Healthcare & Scientific Communications
- BMA, HTN and PMEA (×2) awards, 2025 · for Patiently AI
- Open Targets Hackathon 2025 · Audience Choice
How I work
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.
More at pharmatools.ai/about · Book a call at pharmatools.ai/contact