Research Scientist, User Signals from Agentic Logs

Highlights

• Drive Gemini post-training strategy building alignment flywheel from user signals.
• Build reward and critique models capturing user intent and satisfaction.
• Create horizontal post-training recipes scaling across text, audio, model sizes.

Ideal candidate

Research-driven scientist passionate about aligning advanced models with real users, excited to turn complex interactions and feedback into training signals that improve Gemini.

Requirements

• PhD in Machine Learning, NLP, or related field or equivalent practical experience
• Experience with software development in Python and deep learning frameworks like PyTorch or JAX
• Research experience in post-training, RL, agentic RL, preference learning, or reward modeling
• First-author or co-author publication in peer-reviewed ML/NLP conferences like NeurIPS, ICML, ICLR, ACL, EMNLP
• First-author publications at top ML/NLP venues
• Experience with alignment and post-training methods such as RLHF, DPO, PPO, rejection sampling, or self-play
• Experience modeling multi-turn interaction data, agent trajectories/logs, or implicit and explicit user feedback
• Experience training or evaluating reward models, critique models, or automated evaluators like LLM-as-a-judge
• Experience in agentic post-training