Role Description
**Introduction**
:
* Join us at Fastino as we build the next generation of LLMs. Our team, boasting alumni from Google Research, Apple, Stanford, and Cambridge is on a mission to develop specialized, efficient AI.
* Fastino's GLiNER family of open source models has been downloaded more than 5 million times and is used by companies such as NVIDIA, Meta, and Airbnb
* Fastino has raised $25M (as featured in TechCrunch) through our seed round and is backed by leading investors including Microsoft, Khosla Ventures, Insight Partners, Github CEO Thomas Dohmke, Docker CEO Scott Johnston, and others.
**What You’ll Work On**
:
* Explore novel architectures for small language models, with a focus on improving factual grounding and reducing hallucinations at the architectural level
* Prototype and benchmark new model designs, attention mechanisms, and training strategies in PyTorch
* Design and run rigorous experiments to evaluate factual accuracy, calibration, and failure modes across a variety of benchmarks and real-world tasks
* Collaborate closely with senior researchers and engineers to translate promising research directions into production-ready components
* Contribute to Fastino's research output through internal reports, and potentially author publications at top-tier venues (ACL, EMNLP, ICLR)
**What We’re Looking For**
:
* Currently pursuing or completing a Master's degree (stage de fin d'études) or PhD in Computer Science, Machine Learning, Applied Mathematics, or a related field at a French or European university
* Hands-on experience implementing model architectures in PyTorch, demonstrated through academic projects, personal work, or open-source contributions
* Strong foundations in deep learning and NLP, with solid knowledge of core building blocks: transformers, attention mechanisms, diffusion models, autoregressive models, optimization, and generative modeling
* Genuine curiosity about what makes language models hallucinate and how architecture design can solve it
* Fluent in both French and English (written and spoken)
* Autonomous and rigorous, comfortable working independently, designing experiments, and communicating results clearly
**Nice to have**
:
* (Optional) Publications or preprints in top ML/NLP conferences
* (Optional) Contributions to well-known open-source ML projects
**What we offer**
:
* The opportunity to shape the architecture of models used by thousands of developers and major tech companies worldwide
* A fast-moving startup environment
* Competitive internship compensation
* Flexibility: Remote (France)