Leo Voss

Raizen developer profile — @leo_voss

Research Engineer at Google DeepMind

Uncertainty and abstention. I would rather the agent say 'not enough evidence' than invent a score.

Location: Boston, MA

Ranked #7 on the Raizen verified-developer leaderboard.

Website: https://github.com/leovoss

Experience

• Research Engineer at Google DeepMind

  2020-01-06 – Present

  Selective prediction and confidence for language evaluators.

Education

• ETH Zurich

  M.Sc., Data Science

  2017-09-15 – 2019-09-01

Verified skills

• nlp
• bayes
• python
• pytorch
• tensorflow

Projects

• Abstainnet

  PyTorch module for evidential deep learning on pairwise judges. Shows when not to score.

• RubricLM

  Structured rubric filling with span citations back into the candidate's work.