Head of Population Simulation Research

Location: New York City · In person, five days a week
Compensation: $425,000–$525,000 base

Aaru builds simulations of human behavior. Each simulation contains a population of agents, each representing a person who could plausibly exist in the real world and capable of making decisions within a modeled environment. Companies and institutions use these simulations to test consequential choices before committing—from product launches and policy changes to critical communications. Because the agents are simulated rather than recruited, they can reason through complex hypotheticals without fatigue or the response effects common in human studies.

Simulation research focuses on expanding what simulations can be as well as documenting and verifying that the systems today are worthy of trust. As the Head of Simulation Research, you will set Aaru’s research strategy in collaboration with the founders, conduct original research, and lead a team of researchers.

This is a hands-on research leadership role. Someone successful will be a player-coach; they’ll be able to get down in the weeds of day-to-day IC work alongside strategizing for long term plans. You will choose the bets to make, the way experiments are designed, and recruit exceptional people to join. While research at Aaru is exploratory, it’s not unconstrained – good research must eventually change what Aaru can build and measure.

• →

  Define and continuously refine a coherent research agenda for individual behavior, group dynamics, and population-level validity.

• →

  Choose a small number of high-leverage bets and establish the experiments, baselines, and decision criteria needed to evaluate them.

• →

  Personally design and run experiments, build prototypes, analyze model behavior, and turn ambiguous findings into the next testable question.

• →

  Develop new methods for agent construction, memory, interaction, behavioral calibration, population synthesis, and uncertainty estimation.

• →

  Build evaluation frameworks that compare simulated behavior with real observations and reveal where apparent success does not generalize.

• →

  Establish high standards for reproducibility, falsifiability, negative results, and honest communication of limitations.

• →

  Partner closely with Simulation Engineering. Hand off validated methods with evidence, implementation insight, known failure modes, and clear production hypotheses.

• →

  Hire, mentor, and lead a small team of exceptional researchers and research engineers.

• Build a longitudinal evaluation that tests whether an agent remains behaviorally coherent across changing contexts rather than merely sounding consistent.

• Reconstruct a real-world decision or event, preregister the expected measurements, and test whether simulated subgroups reproduce observed outcomes.

• Design controlled experiments on information diffusion, coordination, polarization, or social influence inside an agent population.

• Develop methods that calibrate confidence and teach a simulation to surface when it should not be trusted.

• Study how fidelity changes with population size, heterogeneity, interaction topology, model capability, and computation.

• Identify a result that looks compelling in aggregate but fails for a meaningful subgroup, then explain the failure and derive a better method.

We believe that simulation is an empirical science. Like other sciences, we value falsifiable hypotheses, strong baselines, careful measurement, reproducibility, and negative results. Our goal is to make a small number of large bets, putting lots of care towards identifying which ones to make and how to execute on them. While papers and benchmarks are useful, the goal–understanding how to make simulation populations as accurate as real people–is the core focus.

• You have developed an original research agenda at a frontier ML lab or an environment with an equivalent bar for rigor and ambition.

• You are equally comfortable forming a research strategy, designing an experiment, writing code, and inspecting individual failures.

• You have led researchers or a major technical direction while remaining a direct contributor to the work.

• You can turn an important but poorly specified question into a sequence of decisive experiments.

• You care more about discovering the truth than preserving an elegant hypothesis or producing a persuasive demo.

• You can synthesize ideas across machine learning, behavioral science, statistics, economics, or complex systems without becoming trapped by one discipline’s conventions.

• You communicate results plainly to researchers, engineers, customers, and company leadership—including negative results and inconvenient limits.

• You want to build in person, in New York, at high speed.

• Work in multi-agent systems, LLM behavior, post-training, evaluation, synthetic environments, or model-based experimentation.

• Work in computational social science, behavioral modeling, economics, psychometrics, causal inference, or adjacent fields.

• A record of research that changed a model capability, product decision, or scientific understanding—not merely a benchmark score.

• Experience recruiting and mentoring unusually strong researchers and research engineers.

• Aaru has a focused research agenda organized around a small number of consequential, testable questions.

• New methods produce measurable improvements in behavioral fidelity or reveal clear limits that change the company’s direction.

• Research evaluations become the trusted basis for deciding what Aaru should build and ship.

• Simulation Engineering receives methods that are reproducible, well-understood, and accompanied by honest evidence.

• A small, exceptional research team develops a reputation for work that is both scientifically serious and useful in the world.