Senior ML Engineer, Model Lifecycle & Spatial AI in München

Own how models are built, trained and proven, from registry to factory edge.

AI Futures is partnering with a Munich-based Physical AI scale-up building smart-glasses-powered AI for the factory floor. Their system helps major automotive OEMs understand what happens on their assembly lines, and their clients accept it through formal tests with hard recall and false-positive gates. That means the models have to provably work: which weights, trained on which data, with which config, always answerable.

We're looking for a Senior ML Engineer to own that machinery end-to-end, and with it two of the company's hardest ML problems.

WHAT YOU'LL DO

• Build the machinery that makes models provable: training pipelines, experiment tracking and a model registry, with full lineage from dataset to deployed weights, plus the evaluation harnesses client acceptance rests on
• Solve data scarcity: build simulation-based synthetic data pipelines for rare anomaly classes that real factories are (thankfully) too good to produce often
• Take the system beyond vision: productionize the audio modality and develop worker-vehicle association, using SLAM fused with vehicle identity signals to know which car a worker is acting on as they move along the line
• Ship at the edge: own the anonymization models the company's privacy guarantees depend on, and optimize everything for constrained GPU/CPU budgets on factory hardware
• Work as a peer of the Lead ML Engineer, who owns what the system should do while you own how models get built, trained and reproduced. You'll design together, in the open, with a direct line to the CTO

WHAT YOU BRING

• 5+ years building and shipping ML in production, with ownership of training pipelines, experiment tracking and model registries (not just using them)
• Strong computer vision background and solid PyTorch skills
• Experience deploying models to constrained hardware (e.g. ONNX, TensorRT, quantization, Jetson-class devices)
• A rigorous approach to evaluation: you know the difference between a good metric and a defensible one
• Comfort setting standards as the person who decides how things are done

NICE TO HAVE

• Synthetic data or simulation experience (Omniverse, Unity, Unreal, Blender, domain randomization)
• SLAM, visual-inertial odometry or sensor fusion
• Privacy or anonymization models
• Background in automotive, robotics, AR/VR or other validation-heavy environments

WHAT'S ON OFFER

• The ML production culture of a company, shaped by you from the start: registry, tracking and evals, your way
• Multimodal problems (vision, audio, spatial) most teams only get one of, on data nobody else has

WHERE YOU'LL BE IN 12 MONTHS

Every model that passes client acceptance is reproducible from the registry. A rare-anomaly class has hit its recall gate on synthetic data. Audio is live in a deployment, and SLAM-based worker-vehicle association is validated on a real line.

#J-18808-Ljbffr