Design and operate large-scale multilingual data pipelines — sourcing, cleaning, deduplication, language identification, and script normalization across high- and low resource languages.
Define and enforce quality bars for multilingual corpora, including translation quality, cultural fidelity, toxicity, and contamination checks.
Design and run scientific experiments to advance our understanding of scaling large language models to improve multilingual data efficiency.
Lead small research projects independently while collaborating on larger initiatives.
Build evaluation sets and diagnostics that expose where model behavior degrades by language, register, or domain, and close those gaps with targeted data.
Work with pre-training, mid-training, and post-training teams to land measurable, step-function improvements in multilingual capability.
Qualification
A rigorousPublished work in multilingual NLP
Required
Strong software engineering fundamentals and comfort processing web-scale datasets in distributed environments.
Experience building large-scale data pipelines for language models, machine translation, speech, or search — ideally covering more than one language.
Fluency or working proficiency in at least one language other than English, and genuine curiosity about how languages differ.
A rigorous, measurement-first mindset: you prove a data change helped rather than assuming it did.
Navigate trade-offs between research objectives and practical engineering realities.
Comfortable with ambiguity and high ownership in a small, fast-moving team. Passionate about advancing the frontier of intelligence.
Preferred
Graduate degree (MS or PhD) in Computer Science, Machine Learning, or related discipline.
Published work in multilingual NLP, low-resource languages, or evaluation methodology.
Experience with annotation vendor management or crowdsourced data operations at scale.
Familiarity with tokenizer design and its effects on non-Latin scripts.
