1 day ago
London, United KingdomMid Level
Responsibilities
- Post-train and iterate GAIA-class world models for rig transfer, pose transfer, and related geometry, calibration, and action conditioning.
- Own the synthetic-data generation loop from configuration through large-scale GPU inference and training-ready artefacts with clear model and settings lineage.
- Integrate synthetic data into behaviour-cloning, reward-model, and reinforcement-learning training, including mix-ratio, quality-filter, ablation, and impact studies.
- Diagnose geometry, calibration, controllability, novel-view-synthesis, odometry, and camera-layout failures in generated video.
- Improve inference throughput, valid-generation rate, and self-serve workflows using techniques such as shortcutting, distillation, KV caching, and reduced sampling steps.
- Expand synthetic-data coverage to new vehicle platforms and safety-critical scenarios including Emergency Lane Keeping and Automatic Emergency Braking.
- Collaborate with world-model researchers, infrastructure engineers, and driving-model owners to connect generation, evaluation, and training.
Requirements
- At least 4 years of applied machine learning or research-engineering experience training and shipping neural networks.
- Strong Python and PyTorch or equivalent experience, including GPU training, debugging, and reading model code.
- Hands-on experience with video, generative, or world models, such as diffusion, flow-matching, autoregressive video, novel-view synthesis, or neural rendering.
- Working knowledge of cameras and 3D geometry, including multi-camera rigs, intrinsics, extrinsics, warps, and reprojection.
- Experience taking generated or simulated data into a downstream trained model and measuring impact through mixing, ablations, and failure analysis.
- Experience operating multi-GPU generation or training jobs, workflow orchestration, and large video artefacts at scale.
- Experience collaborating with researchers and platform engineers while owning an ML capability.
- Desirable experience with controllable generation, distillation, few-step sampling, KV caching, AV, robotics, simulation, multi-sensor driving data, production research workflows, reward models, offline reinforcement learning, closed-loop driving-policy evaluation, cloud GPU fleets, or distributed training.
Benefits
- Full-time position based in London with a hybrid working policy combining office and workshop time with work from home.
- Core working hours with flexibility to determine a schedule that works for the individual and team.
- Opportunity to work on generative simulation, world models, autonomous driving, and models deployed in real vehicles.
- Inclusive interview process with accommodations or adjustments available on request.
Categories
