Specialists in behavioural AI for health, safety, and real-world impact

I have spent my career on one problem: how to design programmes and products that help people change and sustain the behaviours that matter.

That work started in the clinic and in research. I co-founded MEND, a family-based obesity programme that grew to support more than 130,000 families worldwide, proven efficacious through multiple randomised controlled trials.

From there I moved into large-scale digital and AI leadership. I led the digital transformation and international expansion of Slimming World, one of the world's largest weight-management services, whose digital programme grew past a million members. At Allurion, as Global VP of Behavioural Science, I led the behavioural science team behind two of the first large-scale generative AI conversational health coaches in obesity care, and helped build the full provider and patient platform behind it. I have since advised companies in the UK and US on scaling behaviourally informed, AI-driven care.

Alongside the applied work, I publish peer-reviewed research on AI evaluation in high-stakes settings. I recently led a Wellcome Open Research open letter on behavioural science in AI, co-authored with colleagues at UCL, Harvard, Duke, Imperial, and the Alan Turing Institute, and I am a co-author of the FAST framework in Frontiers in Digital Health, a peer-reviewed framework for evaluating conversational AI health coaches.

At Sacher AI, we design and deploy behavioural AI systems for patients, clinicians, and care teams in real-world digital health, across obesity, metabolic health, and mental health.

Much of our current work is in GLP-1 care. The medications are effective for many patients, but the challenge is everything around them: keeping patients engaged, managing expectations and side effects, and supporting the behaviour change that determines long-term outcomes. More than half of patients discontinue within 12 months, even in well-resourced clinical settings. That is not a medication problem. It is a behavioural one.

What makes our approach different is that it is grounded in how people actually think and behave. We use cognitive and behavioural frameworks, including behavioural phenotyping, to understand why individual patients are likely to struggle, and design support that responds to those differences: the same pathway, delivered differently to different people. I am a named inventor on a pending US patent for digital phenotyping and personalised behavioural interventions in obesity.

Evaluation and monitoring are core to the work. We build AI systems, and we help teams see how they are performing and where the risks lie. PromptSafe is our tool for pre-deployment AI evaluation, and we are developing OVRSI for real-time monitoring of patient-facing AI.