Probabilistic AI Validation
I know where LLMs break because I'm the one who broke them.
Bridging the gap between probabilistic AI capabilities and GxP compliance.
For the past several years I've been inside the training loops of frontier models — executing high-complexity RLHF for scientific domains, finding the failure modes, and teaching the models to do better. Before that, I was a Validation Engineer at Kite Pharma (Gilead), ensuring GMP compliance for cell therapy manufacturing.
I bridge the gap between the data scientists/ML engineers and the quality team.
Most consultants offer generic AI strategy. I offer technical assurance. I help life sciences directors navigate the tension between innovation and regulation. I don't just tell you a model works — I provide the audit-ready evidence required to deploy it in a regulated environment.
Research, then industry, then the models themselves
Research
Published scientist, cover of the Journal of Experimental Biology. Specialized in computational analysis of fMRI data — applying statistical rigor to noisy biological signals long before "AI" was a buzzword.
Validation Engineer, Kite Pharma (Gilead)
GMP compliance and regulatory adherence for cell therapy manufacturing. The inspection-side view of what evidence has to survive.
Scientific RLHF specialist
High-complexity reinforcement learning from human feedback for scientific domains. I've trained the models your team is probably evaluating right now.
M.S. Physiology & Biophysics
The quantitative biology grounding underneath all of it — which is why the frameworks read as science rather than compliance boilerplate.
What I do, and what it gets you
Core expertise
- Scientific RLHF Tuning models for accuracy and reasoning, not just fluency.
- GAMP 5 / CSA validation Translating black-box behavior into inspection-ready documentation.
- Adversarial stress-testing Proactively finding failure modes — sycophancy, data lineage breaks — before Quality Assurance does.
What I solve for you
- Accelerate adoption Unblocking pilots stalled by Quality or Legal concerns.
- De-risk deployment Validating that your agents respect SOPs and data boundaries.
- Industrialize AI Moving you from cool demo to production-grade asset.
Who I work with
I focus on pharma and biotech, but also support health-tech and research teams where regulated thinking adds value.
If you're deploying AI in a context where being wrong has consequences, we should talk.
Book a scoping call