Data Engineer: Multimodal ML Data Pipelines

AI Chopping Block; Paris; Hybride; EUR 70 000 - 110 000; Plein temps

Résumé du poste

Harmattan AI is expanding its data engineering team in Paris, Lausanne, or Zurich to own the data layer powering our autonomous defense ML models. You will manage terabytes of raw data, turning it into clean, versioned training-ready datasets so modelers can focus on model design and training.

You will drive ingestion, curation, labeling workflows, dataset construction, and data lineage to ensure reproducibility and efficient model training in a high-responsibility environment.

Qualifications

• Degree in a STEM field or equivalent practical experience.
• Experience building and maintaining pipelines for unstructured data at scale (video, images, sensor data).
• Strong Python and data engineering skills; optimize data loaders for training frameworks (e.g., PyTorch).
• Bonus: multimodal sensor data, dataset versioning tooling, and distributed data processing tooling.
• Attributes: systematic, quality-minded, pragmatic, service-oriented to enable modelers.

Responsabilités

• Ingestion pipeline: ingest, decode, and store raw unstructured field data into efficient formats.
• Multimodal alignment: align data streams temporally and spatially for training.
• Curation: transform ingested and public data into high-value datasets.
• Data & labeling requirements: define data to gather and labeling workflows; coordinate with data teams.
• Dataset construction: build task-specific datasets for training and evaluation with cross-team collaboration.
• Versioning & lineage: version datasets and maintain lineage for reproducible training runs.
• Storage & formats: store data in training-ready formats and manage storage tiering.
• Efficient delivery: provide clean datasets and tools for loading to avoid I/O bottlenecks.

Connaissances

Python

Data engineering

Data pipelines

Automation

Formation

STEM degree

Description du poste

Harmattan AI is expanding its data engineering team in Paris, Lausanne, or Zurich to own the data layer powering our autonomous defense ML models. You will manage terabytes of raw data, turning it into clean, versioned training-ready datasets so modelers can focus on model design and training.

You will drive ingestion, curation, labeling workflows, dataset construction, and data lineage to ensure reproducibility and efficient model training in a high-responsibility environment.