About the role
You make research results and model releases traceable to the data and decisions that produced them. Most irreproducibility is a data-lineage problem rather than a modelling one, and this role is the infrastructure that makes learning responsibly possible rather than aspirational.
The work
Version datasets, model artifacts, experiment configurations and evaluation results. Build provenance, quality checks and reproducible pipelines. Separate household data from shared research datasets, and coordinate deletion and retention behavior with privacy engineers.
What good looks like
In your first 90 days, deliver a reproducible experiment pipeline with dataset lineage and release checks that block unapproved inputs.
Evidence we look for
Bring experience with data systems, ML operations or scientific computing. Show how you have detected contamination, data drift or irreproducible results.
What we need to see
- Data systems, ML operations, or scientific computing experience
- You have detected contamination, data drift, or an irreproducible result, and can describe how
- You build lineage that survives people being in a hurry
- You treat evaluation data as something to protect from leakage
Nice to have
- Experiment tracking and model registry systems
- Data versioning at scale
- You have built infrastructure researchers voluntarily used
The exercise
Trace a model-quality regression after a dataset update and explain which artifacts are necessary to reproduce it.
Where and how we work
In the office together five days a week, in any of these cities. Remote-friendly around your family, arranged one person at a time.