All roles
Engineering · London

Applied AI Engineer

Build reliable AI systems that do real operational work inside private clinics, with clear boundaries, human approval, and evidence for every action.

London · Hybrid · Full-time
Apply for this role

Why this role exists

Clinics do not need another chatbot. They need dependable machinery that can notice missing work across fragmented systems, prepare the right next action, and stop whenever judgement is required.

You will turn that problem into production software. The role spans agents, workflow orchestration, integrations, evals, audit trails, and the small product details that make a clinic team trust what is running.

What you will do

  • Build and operate workflow services across patient management systems, messaging tools, forms, email, and clinic data sources.
  • Design AI-assisted steps with deterministic boundaries, typed outputs, explicit human gates, and safe exception paths.
  • Create eval suites and synthetic clinic scenarios that test matching, routing, drafting, duplicate handling, and failure recovery.
  • Make every action explainable through useful logs, replayable events, versioned rules, pause controls, and operational dashboards.
  • Work directly with implementation and clinic teams to understand edge cases before they become incidents.
  • Own features from discovery through deployment, monitoring, iteration, and measurable clinic outcomes.

You will probably thrive here if

  • You have shipped production backend, full-stack, data, or automation systems and can make sensible architectural trade-offs without a committee.
  • You are comfortable in TypeScript or Python and understand APIs, queues, databases, observability, and testing deeply enough to debug the whole path.
  • You have built with modern language models and know that prompting is the start of the work, not the end. Evals, constraints, and fallbacks are normal engineering tools to you.
  • You care about the user at the end of the system. You can sit with a practice manager, understand why the workflow fails on Tuesdays, and turn that into a better product.

What good looks like

  • Within 30 days, you can trace a live clinic workflow end to end and improve its tests or observability.
  • Within 90 days, you have shipped a meaningful workflow change into controlled production and can show what changed for the clinic.
  • Within six months, you own a reliable part of the operating layer and have made it easier to launch the next clinic safely.
Interested?

Send the work that best shows how you think.

A CV is useful. A concise note about a relevant system, implementation, clinic problem, or decision you owned is better.

Apply for this role