We're building AI to predict clinical trial and drug development outcomes, and quantify how much confidence those predictions deserve.
We're looking for a technical co-founder to join our biology and commercial founders. As Co-founder & CTO you take a built, already-used platform and lead its development, the research programme that measures and improves its predictions, and the first technical team. The aim is to spin the company out of DSV over the coming months.
THE OPPORTUNITY
Drug development involves decisions about targets, treatment approaches, delivery and patient populations. Each decision depends on evidence that may not transfer well to the setting that matters: treating people. A treatment working in mice, for example, does not tell us how much confidence to place in its chances in humans.
Around nine in ten drugs entering clinical trials never reach approval. The causes run from efficacy to safety to commercial choice, and the claim here is narrower than blaming that number on translation: the field still lacks a systematic way to measure how far a given experimental result should carry into a specific therapeutic decision.
We want to measure how reliably different kinds of biological evidence predict human outcomes. Our approach is to link the evidence available when a decision was made, the judgement made from it, and what happened next. Those records become the data from which we can learn which evidence predicts which outcomes, and under what conditions. That is difficult to test. Outcomes can take years to arrive, suitable labels are often missing, and a failed trial does not necessarily reveal which earlier assumption was wrong.
The platform is already running. It has produced a scientific result through work with the Allen Institute, described in a public preprint . You can also see the engine run .
There is no established dataset or method that settles the measurement question. Defining what to measure, building reliable evaluations and choosing appropriate methods are central to the role, and the next funding round will fund that programme.
Clinical trial outcome prediction is a central test: predicting a specific trial outcome and explaining the biological reasons, using evidence available before the result. An LLM may already know a historical outcome, so retrospective tests need safeguards against contamination as well as confirmation on prospective cases. This article sets out our approach.
The aim is to make therapeutic development more predictable, and use those predictions to change which treatments get developed and how.
Read more in our thesis .
THE ROLE
This is a deeply hands-on role. The founding team covers biology and commercial development. You will be responsible for the technology and the research needed to evaluate it. There is no engineering team yet: you will be building the system yourself and should expect to remain hands-on for at least the first year. You will:
Develop the platform. Take ownership of the existing system, set technical priorities and implement improvements.
Build the measurement programme. Define the prediction tasks, link evidence available at decision time to subsequent outcomes, and develop and compare modelling approaches. Test predictive performance and calibration, and identify where the system is reliable and where it is not.
Prevent misleading evaluations. Address leakage, label quality and contamination. Retrospective tests need particular care when an LLM may have encountered the outcome during training. We explain the issue in this article .
Hire the first technical team once funding is in place.
Support fundraising. Explain the technical approach, evidence, limitations and development plan to investors.
The modelling approach is open. You will choose methods based on the problem and the available data, whether those involve Bayesian or hierarchical inference, graph learning, calibration, classical statistics or a combination.
Location: Flexible, UK-based…