About the teamZillow has the richest housing market data in the industry. You help us build out a foundational model that leverages a representation of who is in the market, what they'll pay, and how that demand clears against supply.
The Agentic Foundations team is a focused group of 15-20 applied scientists within Zillow's broader Agentic AI organization, building the foundational AI technologies that power the next generation of home shopping experiences. This Principal Applied Scientist will help build the Zillow World Model: a single, uncertainty-aware model of the housing market that ties together supply, demand and macro conditions such as mortgage rates and policy. Today, bespoke models like Zestimate, Time-to-Pending and Listing Soon each answer one question well. The foundational "World" Model aims to answer the multi-step "what if" questions consumers actually ask, such as "If I list at $500K, when will it sell?" or "I want at least $800K; what should I list at?", as consistent projections of one market state, and to serve those answers to Voyager, Zillow's AI Mode, as a tool.
This is a very new field and an open-ended research effort. You will partner with the World Model working group (listing-state modeling, market simulation and evaluation) and with product, engineering, platform and data teams. Zillow observes both sides of the housing market at a depth no one else can, so this work can directly help guide users toward their goals.
About the role- Build the demand layer of the World Model: turn large-scale shopper behavior (search, engagement, session trajectories) into structured demand representations, including explicit attention-to-demand conversion.
- Design and train latent state-space, multi-tower and/or sequential dynamics models with calibrated uncertainty that support multi-step rollouts, forward planning ("if I do X, what happens?") and backward planning ("I want X; what should I do?").
- Estimate willingness-to-pay, per-home demand curves and market clearing dynamics, connecting learned representations to economic quantities.
- Apply causal identification in observational data and define where the model must abstain because a correlation does not support a "what-if" claim.
- Contribute to the World Model evaluation battery: state probes, calibrated multi-target forecasts, rollout drift, causal checks and out-of-regime replays.
- Independently drive ambiguous research end-to-end, and provide technical leadership and mentorship in translating state-of-the-art research into production impact.
This role has been categorized as a Remote position. "Remote" employees do not have a permanent corporate office workplace and, instead, work from a physical location of their choice, which must be identified to the Company. U.S. employees may live in any of the 50 United States, with limited exceptions.
In California, Connecticut, Maryland, Massachusetts, New Jersey, New York, Washington state, and Washington DC the standard base pay range for this role is $195,600.00 - $312,400.00 annually. This base pay range is specific to these locations and may not be applicable to other locations.In Colorado, Hawaii, Illinois, Maine, Minnesota, Nevada, Ohio, Rhode Island, Vermont, and Virginia the standard base pay range for this role is $185,800.00 - $296,800.00 annually. The base pay range is specific to these locations and may not be applicable to other locations.
In addition to a competitive base salary this position is also eligible for equity awards based on factors such as experience, performance and location. Actual amounts will vary depending on experience, performance and location. Employees in this role will not be paid below the salary threshold for exempt employees in the state where they reside.
Who you areMust-haves- PhD in Computer Science, Machine Learning, Statistics, Economics, or a related quantitative field.
- Hands-on experience building latent state-space or world models (RSSM/Dreamer-family, JEPA-style latent prediction, or event-sequence transformers): belief-state representations, stochastic transition models with calibrated uncertainty, and evaluation of multi-step rollouts, not just single-target forecasting.
- Demonstrated ability to turn large-scale user behavior data into structured demand representations (utility or willingness-to-pay estimation, funnel/conversion modeling) while correcting for selection bias in platform-observed populations.
- Practical causal inference in observational settings (quasi-experiments, instruments, policy discontinuities), and judgment about when a learned correlation does and does not support a counterfactual claim.
- Proven record of delivering on open-ended research projects, with the ability to scope ambiguous problems and provide technical leadership to scientists and MLEs.
- Strong programming skills, especially Python, plus experience with ML frameworks such as PyTorch or TensorFlow.
Nice-to-haves- Economics fluency (strongly preferred): market microstructure, search-and-matching, or structural/discrete-choice demand models; reasoning about equilibrium feedback.
- Experience with agent-based or market simulation, or simulation-based reinforcement learning.
- Published work in world models, model-based RL, causal inference or structural demand estimation.
- Experience with multimodal representation learning or bitemporal, leakage-safe training data.
- Familiarity with two-sided marketplaces, real estate, or long-horizon consumer journeys.
