AI Researcher – Multilingual Data
Featherless AI · Remote (world)
Extracted from the posting
Works on data sourcing, filtering, deduplication and evaluation for multilingual and low-resource language model training. Builds multilingual benchmarks and researches cross-lingual transfer and alignment. Stack is Python with PyTorch or JAX.
About the Role
We’re looking for an AI Researcher focused on multilingual data to help us build and scale next-generation language models across diverse languages and domains. You’ll own research and execution around data sourcing, curation, evaluation, and training strategies for multilingual and low-resource languages, with a strong emphasis on publishing high-quality research and translating it into production systems.
This role is ideal for someone who enjoys working close to the frontier: balancing papers, prototypes, and real-world impact in a fast-moving startup environment.
What You’ll Do
- Design and execute research on multilingual datasets, including data collection, filtering, deduplication, and quality measurement
- Develop strategies for low-resource and long-tail languages (sampling, augmentation, curriculum design)
- Research and improve cross-lingual transfer, alignment, and robustness in large language models
- Build and maintain evaluation benchmarks for multilingual performance
- Collaborate with engineers and researchers on training pipelines and model architecture decisions
- Publish research at top venues (e.g., ACL, EMNLP, NeurIPS, ICML, ICLR) and contribute to open-source when appropriate
- Translate research insights into practical improvements in production models
What We’re Looking For
- Strong background in NLP / ML research, with a focus on multilingual or cross-lingual modeling
- Publication record at respected conferences or journals (ACL, EMNLP, NeurIPS, ICML, ICLR, etc.)
- Experience working with large-scale text datasets across multiple languages
- Solid understanding of:Tokenization and vocabulary design for multilingual modelsData quality metrics, filtering, and dataset biasTransfer learning and multilingual representation learning
- Comfortable prototyping in Python with modern ML frameworks (PyTorch, JAX, etc.)
- Ability to operate independently and ship research in a startup pace environment
Nice to Have
- Experience with low-resource languages or non-Latin scripts
- Open-source contributions in NLP or data tooling
- Experience training or evaluating large language models
- Familiarity with multilingual benchmarks (e.g., XTREME, FLORES, TyDi QA)
Why Join Us
- Real ownership over research direction and impact
- A team that values papers and production
- Access to meaningful scale: large datasets, modern infrastructure, and fast iteration
- Competitive compensation and meaningful equity at an early stage
