Applied Research Scientist, World Model

Role overview Help develop a unified, uncertainty-aware model of the housing market that connects shopper demand, available homes, and broader market conditions. The research aims to support multi-step questions—such as how a change in asking price could affect sale timing—by producing consistent projections that can inform an AI-assisted home-shopping experience. The role centers on modeling demand from large-scale behavior data and translating open-ended research into practical systems.

Responsibilities - Build structured representations of shopper demand from searches, engagement, and session sequences, including how attention relates to demand. - Design and train latent state, multi-tower, or sequential models that support planning and multi-step prediction with calibrated uncertainty. - Estimate willingness to pay, home-level demand curves, and market-clearing behavior. - Use causal methods to assess what-if questions and define when evidence is too weak for a model to make a claim. - Develop evaluation methods covering model state, forecast calibration, rollout drift, causal validity, and performance outside familiar conditions. - Lead ambiguous research work independently, mentor colleagues, and partner across science, product, engineering, platform, and data teams.

Requirements - PhD in computer science, machine learning, statistics, economics, or a related quantitative field. - Hands-on experience with latent state-space or world models, including stochastic transitions and uncertainty calibration. - Experience turning large-scale user behavior into structured demand or preference representations. - Ability to evaluate multi-step model rollouts, beyond single-target forecasting. - Research and technical leadership skills for taking open-ended work toward production use.

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