Visiting Scholar — Post Training & Research (Discovery AI Lab)

About the job

Discovery AI Lab is pioneering large language model (LLM) technologies across the full ML lifecycle — pre-training, mid-training, and post-training — to replace traditional recommendation pipelines with generative AI. Our work spans preference alignment and reinforcement learning for recommendation systems, self-improving agentic AI, and efficient training and inference at scale. We publish at top venues (NeurIPS, ICML, ICLR, RecSys) while shipping research directly into products that serve billions of users. We are seeking a Visiting Scholar to join the team for a 12-month, full-time research engagement. This role offers the opportunity to lead cutting-edge research embedded within a world-class team, with access to Meta-scale infrastructure, data, and compute.

Responsibilities

Lead research on post-training algorithms for generative recommendation systems, including preference alignment methods (e.g., DPO, GRPO, SimPO) adapted for multi-objective recommendation signals. Design and develop self-improving agent frameworks that leverage multi-agent collaboration, LLM self-correction, and continuous-learning loops. Advance efficient inference techniques — including quantization, compression, and distillation — for large-scale generative and Mixture-of-Experts recommendation models. Collaborate with research scientists and engineers to translate research into production-ready systems at Meta scale. Mentor research scientists and engineers on the team, upleveling internal capabilities in post-training and agentic AI.