About the role
As a Model Behaviour Architect in the Multilingual team, you define and measure what good model behaviour looks like across languages, not just in English. You will work hand in hand with our Science team to define what “good” means for multilingual behaviour, covering language adherence, naturalness, cultural alignment, and consistent capability across languages in which our models are deployed.
What you will do
Interact with models across languages to identify where multilingual behaviour can be improved
Gather internal and external feedback to scope areas for improvement
Define what high-quality behaviour means per language
Design and implement evals, data guidelines, data generation, and synthetic testing environments for multilingual settings
Shape how models reflect the cultural context of the users and regions they serve
Identify and fix edge-case behaviours through rigorous cross-language testing
Develop robust evaluation pipelines for model candidates
Work collaboratively with AI Scientists
About you
You are an expert at building robust evaluations
You have a strong understanding of at least one of: linguistics, language, translation; LLM agents at work, including reasoning and tool use. Depth in linguistics or multilingual NLP is strongly preferred
You thrive in dynamic and technically complex environments
You have a track record of delivering innovative, out-of-the-box solutions to address real-world constraints
#J-18808-Ljbffr
Model Behaviour Architect- Multilingual
Mistral AI · Paris, KY, USA ·
- Job type:
- Full Time