Principal ML & AI Engineer (Spatial AI & Perception)
HERE Technologies2 months ago
Berlin, Germany +3 moreStaff+
Responsibilities
- Develop and train infrastructure perception models for lane detection, road boundaries, traffic signs and lights, and road-surface attributes.
- Build BEV-based scene-representation and occupancy models, integrate geospatial priors, and validate synthetic training data.
- Own model optimization for edge deployment, including ONNX export, TensorRT/QNN compilation, graph optimization, quantisation, compression, and accuracy recovery.
- Profile and benchmark models on embedded hardware and build automated regression pipelines for latency, throughput, memory, and release-target validation.
- Translate state-of-the-art research into reproducible, scalable training pipelines and conduct experiments, A/B tests, and ablation studies.
- Collaborate with research, simulation, and technical leadership; support architecture decisions; maintain code quality and documentation; and mentor junior engineers.
Requirements
- 5–7 years of experience in ML/AI, including 3+ years in computer vision, perception, or deep learning systems.
- Proven experience shipping production perception models into real-world systems or large-scale data pipelines.
- Hands-on experience optimizing and deploying models on embedded hardware or automotive-grade SoCs.
- Strong PyTorch expertise, including custom architectures, loss functions, training loops, DDP/FSDP distributed training, mixed precision, and gradient checkpointing.
- Experience with large-scale distributed training beyond fine-tuning pre-trained models.
- Experience with BEV or multi-camera transformer architectures such as BEVFormer, BEVDet, BEVFusion, or occupancy networks.
- Experience with scalable training infrastructure, data-loading optimization, augmentation pipelines, and experiment tracking tools such as W&B or MLflow.
- Experience with structured scene understanding, semantic or instance segmentation, and BEV-space, occupancy-grid, or map-aligned representations.
- Hands-on expertise with ONNX and TensorRT/QNN, quantisation workflows including QAT and PTQ, mixed-precision inference, per-layer sensitivity analysis, accuracy recovery, and model compression.
- Experience with automotive SoC platforms such as Qualcomm Snapdragon Ride, NVIDIA Orin, or TI TDA4, including real-hardware latency profiling, memory-bandwidth analysis, and inference-throughput optimization.
- Ability to evaluate data coverage, failure modes, edge cases, accuracy-latency trade-offs, and regression performance across model variants.
- Ability to reproduce research papers, adapt architectures, and integrate them into scalable, production-ready training loops.
Benefits
- 30 paid vacation days.
- Hybrid work arrangement with 2 days in the office and 3 days working from home per week.
- Flexible working hours and work-life balance support.
- On-site gym and sauna, yoga room, and wellness programs for the Berlin location.
- German language course for Germany-based employees and a BVG ticket for the Berlin location.
- Opportunity to work on high-scale services serving and storing petabytes of data.
- Employee life-coaching sessions, team events, talks, and other community activities.
- Competitive salary plus bonus.
Tech Stack
MLflowPyTorch
Categories
About HERE Technologies
HERE Technologies builds a location data and mapping platform used by automakers, logistics providers, and enterprises to power navigation, routing, and geospatial analytics. It sells map data, APIs/SDKs, and platform services for fleet, mobility, and supply-chain applications. The company traces its roots to Navteq (founded 1985) and Nokia Maps, and is privately held with headquarters in Eindhoven, Netherlands.