Principal Applied Scientist (User Modeling)

Role overview Develop user models that help AI agents understand a shopper’s progress, goals, and how decisions may affect later outcomes. The work applies sequence modeling, causal inference, and reinforcement learning to complex, long-running customer journeys, with the aim of bringing research into production systems. It involves independent problem ownership and close collaboration with scientific and engineering partners.

Responsibilities - Build and run language-model post-training workflows, including supervised fine-tuning, preference optimization, and reinforcement fine-tuning on GPU infrastructure. - Train reward models that produce scalar scores from comparative preference examples across multiple categories. - Assess live agent trajectories using on-policy and online evaluation methods, including LLM-based judging. - Turn offline evaluation criteria into reward functions that generalize to varied production traces. - Drive ambiguous projects from scoping through delivery with partner teams. - Provide technical leadership and mentorship while connecting current AI research to production needs.

Requirements - PhD in computer science, statistics, economics, machine learning, or a related quantitative field, with depth in reinforcement learning or causal inference. - Experience modeling large-scale user behavior with event, session, or trajectory sequences and predicting intent or long-term outcomes. - Applied knowledge of causal inference or reinforcement learning for sequential decisions, such as treatment-effect estimation or off-policy evaluation. - Ability to scope and solve complex work independently and guide scientists and machine-learning engineers. - Working proficiency in Python and a machine-learning framework such as PyTorch or TensorFlow.

Nice to have - Economics or econometrics experience, including demand or consumer-choice modeling. - Research in reinforcement learning, causal inference, survival analysis, or user modeling; experience with behavioral simulation, rare-event modeling, or probability calibration. - Familiarity with two-sided marketplaces or long-horizon consumer journeys.

Benefits and work setup - Remote within the United States, with some location limitations. - Equity awards are available; compensation varies by location, experience, and performance.