Applied AI Researcher (Multimodal 3D)

Job description

01Health is at an inflection point. The platform is built, revenue is accelerating, and we are moving from a single-specialty company to a multi-vertical specialist healthcare platform, with AI as embedded clinical infrastructure. Our vision is to help all clinicians deliver the latest innovations in healthcare, improving the standard of care for millions of people. Backed by the investors behind Revolut, CityMapper, and Depop - we're scaling fast across the UK and beyond, and are looking for exceptional people to join us on our mission. **What is happening at 01Health right now:** - Balderton-backed Series A company - 90% of UK patients can reach an 01Health–affiliated clinic within 30 minutes - All growth has been inbound, ~50% word of mouth - 4.98★ customer rating from the dentists who use us - Aerox (sleep) ready for national rollout to top-performing 01 Partners, with a 300+ clinic waitlist - US expansion live - New verticals being tested and prototyped We are not searching for product–market fit anymore. We are ready to dominate the market, and to do it across multiple medical specialties, transforming lives in the process. ### The Role Right now, turning a scan of someone's mouth into a precise, staged plan for how their teeth should move is one of the hardest, most manual parts of what we do. It takes real clinical skill, it takes time, and it is one of the biggest constraints on how fast we can grow. We think it's also one of the most interesting applied AI problems in healthcare: teaching a model to genuinely understand 3D dental morphology and written guidelines together, fusing scan and text to reason about how it should change and why. We're hiring an Applied AI Researcher to own the research and applied work behind this problem end to end. This isn't a role where someone else defines the approach and you implement it; you'll be figuring out where to even start, prototyping fast, with an aim to to ship what works quickly. You'll sit at the intersection of 3D geometry processing, computer vision, and multimodal (vision-language) deep learning, working closely with our clinical and orthodontic teams (who bring the domain expertise) and our engineering team (who'll help you get what you build into production). This is a genuinely greenfield problem inside a company that already has real scale, real data, and real urgency to solve it. ### What you'll do - Explore and design multimodal models that fuse 3D scan geometry with clinical text (goals, constraints) to condition and steer model outputs - Build and maintain 3D data pipelines for dental meshes: segmentation, cleaning, registration/alignment, remeshing, and resampling, using tools like Open3D, trimesh, and PyTorch3D/PyTorch Geometric, across potentially multiple formats (STL, PLY, and OBJ) - Build 2D computer vision components for detection and segmentation tasks, from modern approaches (YOLO, Mask R-CNN, etc.) to classic CV (thresholding, keypoint detection, morphology) - Design, train, and iterate on deep learning models spanning CNNs, transformers, and point/graph networks (e.g. PointNet), using supervised, unsupervised, and reinforcement learning approaches, with smart augmentation strategies to make the most of limited labelled data - Own the rigor behind training and evaluation: choosing the right losses, defining and tracking the metrics that matter and making sound architectural calls - Take a "where do you even start?" problem and turn it into something that works end to end, from first prototype to a shipped, production system. - Build the ML engineering foundations that let this move fast: GPU training (including serverless), data and annotation pipelines, and experiment tracking ### What success looks like - A model can take a scan plus natural-language guidelines and produce or steer a clinically valid staged plan - The hardest, most manual part of turning a scan into a treatment plan gets meaningfully faster and more consistent, without compromising clinical quality - Your models move from notebook to production - shipped, used, and trusted by the team (not left as artifacts) - You've built evaluation rigor the whole team can trust: when a model improves, everyone can see why and by how much - You're the go-to person in the company for "can we solve this with vision/3D ML, and how?”. Clinicians and engineers alike come to you with problems, not just solutions - The data and annotation pipelines you've built make it faster for the next model, and the one after that, to get built - You've turned genuine research ambiguity into shipped, working systems ### You'll thrive in this role if… - You've built multimodal or vision-language models, fusing image or geometry encoders with text, and understand the failure modes of cross-modal alignment - You have hands-on experience across 3D geometry processing (meshes/point clouds) and 2D computer vision, and you're comfortable moving between the two - You've trained and shipped deep learning models - CNNs, transformers, and/or point/graph networks, and you understand the trade-offs between supervised, unsupervised, and reinforcement approaches - You care as much about evaluation as you do about modelling. You know that a model without the right metrics is just a guess - You've worked with limited or messy labelled data before and have real strategies (not just hope) for getting good results anyway - You're an applied researcher at heart: you'd rather ship something that works end to end than polish something that never leaves a notebook - You're comfortable with genuine ambiguity. Nobody has solved this exact problem before, and you find that exciting rather than daunting - You have the ML engineering chops to support your own research without needing someone else to operationalise your work for you - Bonus: prior exposure to medical or dental imaging, though we care far more about depth in 3D vision and applied ML than domain-specific experience ### Why this role - You'll own one of the hardest and most valuable unsolved problems in the industry, from first principles to production - You'll work directly with real clinical data and real clinical experts, and see your work translate into something that changes how thousands of people's treatment is planned - You'll have genuine research latitude balanced with the urgency and resourcing of a company that is scaling fast and needs this solved - You'll be building out AI-native visual intelligence here, shaping the tooling, standards, and team that come after you - Clear path to grow as the problem space grows - deepening into a larger research function, or broadening into the wider AI platform as it takes shape - Compensation is benchmarked to strong applied AI research roles in the London market **Why join us:** - **Real impact, every day** - your work directly changes patient outcomes. You'll feel it, and so will the people whose care you're improving. - **25 days holiday + your birthday off** - because no one should work on their birthday! - **Outsized leverage and full ownership** - real authority today, lasting impact tomorrow; what you build now shapes how this company operates at every stage after. - **Grow fast** - backed by top VCs and built by the best; you'll be learning from some of the sharpest minds in health-tech every single day. - **Vitality Health cover** - private medical insurance, because we take care of the people who take care of the business. - **Team Perks** - enjoy regular team lunches, quarterly socials, and an annual company retreat.