← All roles
ResearchZurich (hybrid) or remote
Full-time
Interpretability Engineer
You'll make our models explainable: building the analyses that show clients why a model decides what it does, what it has learned internally, and where it is likely to fail.
RoleInterpretability Engineer
What you'll do
- Apply and adapt attribution methods to new architectures and data types
- Probe what internal units and representations encode
- Design stress tests and failure analyses for models before deployment
- Turn findings into reports and visualisations that non-specialists can act on
What we're looking for
- Solid understanding of deep learning and of interpretability methods
- Experience implementing analyses in PyTorch or JAX
- Scepticism: you check whether an explanation is actually faithful
- Ability to explain technical results clearly to non-experts
Nice to have
- Research experience in interpretability or mechanistic analysis
- Experience with regulated domains such as healthcare or finance
The details
- Location
- Zurich (hybrid) or remote
- Type
- Full-time