Research Scientist - Reality Labs

Research Scientist - Reality Labs Reality Labs at Meta is seeking a Research Scientist with deep expertise in Large Language Models (LLMs) to advance our work in AI-powered neuromotor interactions for wearable devices. We are building next-generation language capabilities that enable natural, expressive, and intelligent communication through Meta's wearable platforms via our industry-leading electromyography (EMG) program. Our team within the Wearables Input and Interaction organization is multi-disciplinary, combining machine learning, language modeling, and on-device optimization to push the boundaries of what's possible with language AI on constrained hardware. This is a unique opportunity to conduct cutting-edge LLM research with direct product impact — shaping how people interact with the next generation of wearable devices. We are looking for a researcher who can drive innovation in language model development, optimization, and deployment while setting technical direction for a small team. If you are passionate about building state-of-the-art language models and translating research into real-world products, we want to hear from you.

Programming experience in Python and hands-on experience with deep learning frameworks such as PyTorchExperience developing machine learning models at scale from inception to impact 5+ years of research experience working autonomously on ML/NLP problems Proven track record of architecting knowledge distillation (KD) pipelines to compress frontier LLMs into optimized edge models Experience bringing LLM-based products from research to production Demonstrated software engineering experience via internship, work experience, or widely used contributions in open source repositories Technical leadership experience setting direction for a team of 3-4 researchers/engineers First-authored publications at peer-reviewed AI/NLP conferences (e.g., ACL, EMNLP, NeurIPS, ICML, ICLR, NAACL) Experience with on-device or edge language model optimization (quantization, sparsity, distillation, knowledge distillation) Deep expertise in logit matching, task-specific SFT data synthesis, and instruction-tuning sub-billion models for production tasks