AI/ML Engineer

About the role

Build the AI therapist that delivers end-to-end CBT to patients in the US: live, clinically supervised, and measured on whether people actually get better.

About Limbic

Limbic’s vision is to ensure the highest quality therapy is accessible to everyone, everywhere.

Limbic has supported more than 820,000 patients across the UK and the US. We are used in over 50% of NHS Talking Therapies services, which makes us the largest-ever deployment of generative AI for direct patient care, and our AI now delivers end-to-end therapy to patients in the US under the supervision of licensed clinicians.

Safety and science come first. We are the only AI mental health chatbot with UKCA Class IIa medical device certification, and the first AI-led mental healthcare company selected for the FDA’s TEMPO pilot. Our results are published in leading scientific journals: in a randomised, double-blind trial published in Nature Medicine, our AI agents scored higher than licensed CBT therapists on a standard measure of therapy skill.

The Role

As an AI/ML Engineer, you will sit at the intersection of computational psychiatry, clinical science, and software engineering. You will join the product team building our AI therapist: a voice agent that delivers structured, evidence-based CBT over the phone to adults with depression and anxiety, with licensed clinicians overseeing every patient’s care.

The product is live and growing week on week, so your work reaches real patients quickly. You will own problems end to end: frame them with clinicians and product managers, ship a change, measure whether it helped, and iterate. Our AI products are built on shared platforms for agents, evaluation and safety, so what you build here is reused across Limbic.

What You’ll Do

- Improve how our AI therapist runs sessions, from session structure and case formulation to therapeutic techniques and continuity across a course of treatment, working closely with clinicians.

- Strengthen the safety systems that protect patients in real time, including risk detection and how the agent responds when risk is disclosed.

- Build evaluations and LLM-as-judge autolabellers, validated against clinician labels, so we know quickly and reliably whether a change made care better or worse.

- Make voice conversations feel natural: responsive, well-paced, and able to recover gracefully when a conversation goes off script.

- Help patients stay with their treatment long enough to recover, through AI features such as well-timed session reminders and clear expectations about what therapy involves, and test what works with experiments.

- Optimise how we collect outcome measures such as PHQ-9 and GAD-7, including the methods, timing and predictive modelling of recovery.

- Ship changes safely through incremental rollouts, automated testing and a clinical sign-off process built for a regulated medical device.

What You’ll Bring