Data/ML Engineer, Fusion Data

The Fusionality Data/ML Engineer joins the founding engineering team that Fusionality is assembling in Lausanne to build its core technology platform. Fusionality is building the measurement, simulation and control stack for the fusion energy reactors of the future. It spun out of EPFL’s Swiss Plasma Center and describes itself as backed by top investors, with a technology that enables reliable operation, real-time monitoring and analysis of high-temperature plasmas in fusion devices.

The Fusionality Data/ML Engineer role suits someone who likes turning large, messy, heterogeneous experimental data into clean, labelled, trustworthy datasets, and who has done this with scientific or industrial sensor data. Familiarity with fusion is a plus but not a strict requirement, and Fusionality’s founders have backgrounds at the intersection of fusion physics, advanced control and AI, including Google DeepMind and EPFL.

Key Responsibilities

Build, maintain and run pipelines that bring experimental fusion data into high-quality, consistent, documented, versioned datasets
Develop ML-assisted labelling and event detection, for example for disruptions, MHD activity and diagnostic faults, using weak supervision, active learning and anomaly detection
Own data quality and provisioning, including provenance, calibration metadata and automated validation, and manage and facilitate access for the rest of the team
Work with physicists validating the datasets and labels
Work with Fusionality’s computational scientists and control engineers, who build models and controllers on these datasets, and with customers to understand and integrate their data

Requirements

Essential:

• Hands-on experience building data pipelines for large scientific or industrial time-series data formats (for example HDF5, Parquet, Zarr)
• Proficiency in designing, querying and optimising SQL and NoSQL database systems, and familiarity with time series databases such as TimescaleDB and InfluxDB
• Experience applying ML techniques to time series in Python (PyTorch or JAX), including sequence models, anomaly and event detection, and semi- or weakly-supervised labelling
• Solid software engineering habits, with testing, reproducibility and data versioning second nature
• Evidence of independent work at the highest level, such as a PhD thesis and/or published code that is used by others, or an industry project you owned, and comfort designing data architectures independently from day 1

Desirable:

• Familiarity with fusion
• Experience with human-in-the-loop labelling tools and workflows
• Experience with pipeline orchestration (for example Dagster, Airflow, Prefect) and data versioning (for example DVC, lakeFS)
• Experience with data from large physics experiments, such as fusion, accelerators or astronomy, or exposure to fusion data systems (MDSplus, IMAS)