Lead Data Engineer

Stellenbeschreibung:

Germany | Full-time, on-site | Competitive six-figure base

We're working with a fast-growing deep tech company in Germany that is developing AI for machines operating in the physical world. As their models scale, so does the volume and complexity of their real-world data, and they're hiring a senior engineer to take ownership of how that data is collected, processed and turned into better models.

This is a hands-on, high-impact role sitting between software and machine learning, with genuine technical authority and a direct line to senior leadership.

The role

Design and run large-scale distributed processing for video, sensor and time-series data
Build the infrastructure that gets data efficiently from collection into large GPU training runs
Work closely with ML researchers to understand where models fall short, and fix it through better data
Define how multimodal data is structured, stored, versioned and quality-checked
Develop Python tooling for validation, cleaning and dataset management
Help shape data collection, annotation processes, and secure handling of sensitive data

What you'll brin

Significant experience in data engineering, ideally with large volumes of real-world data from robotics, autonomous driving, drones or video AI
Strong hands-on experience with distributed processing frameworks such as Ray, Spark, Dask or Beam
Expert, production-grade Python, including asyncio, multiprocessing and performance optimisation
Strong working knowledge of PyTorch and NumPy, and of how training workloads consume data
A proven track record building pipelines that feed multi-node GPU training on AWS or GCP
Experience with high-throughput storage and data loading for training (e.g. S3/GCS, Parquet/Arrow, WebDataset or similar)
Hands-on experience with video encoding (H.264/H.265, FFmpeg) and robotics or sensor log formats (e.g. ROS bags, MCAP or similar)
Dataset versioning and lineage at scale using tools such as DVC, LakeFS, Delta Lake or Iceberg
A solid grasp of machine learning fundamentals, enough to judge which data actually improves a model

Nice to have

Building training datasets for multimodal or foundation models (e.g. VLA, VLM)
Time synchronisation across high-frequency sensor streams and 3D coordinate transforms
Active learning, data-centric ML or failure-case mining workflows
Workflow orchestration (Airflow, Dagster, Prefect) and containerised infrastructure (Docker, Kubernetes)
Experience working with annotation vendors or data collection operations