Population Simulation Researcher

Aaru•New York, NY

3d•Onsite

About The Position

As a Researcher at Aaru, you will design, build, and improve the simulation systems that generate high-fidelity synthetic agents from real-world profiles. The Research team is responsible for Aaru's core simulation capabilities — the technology that enables our customers to test decisions against realistic populations before committing to them.

Requirements

Deep experience with synthetic data generation, agent-based modeling, computational social science, or population simulation
Interest in using LLMs as a simulation substrate — not just for text generation, but for producing structured behavioral outputs that hold up under statistical scrutiny
Ability to think carefully about what it means for a simulation to be "accurate" and can articulate the difference between individual-level plausibility and distributional fidelity
PhD in ML, computational social science, cognitive science, statistics, psychometrics, or a quantitative field — or equivalent depth from industry research
Ability to communicate clearly about what's working and what isn't, including when results are null or ambiguous
Ability to move fast and are comfortable producing work at high velocity without sacrificing rigor

Nice To Haves

Built systems that generate or evaluate synthetic populations at scale
Experience with survey methodology, psychometrics, or demographic modeling
Published work on agent-based simulation, synthetic data, or LLM-based behavioral modeling
Worked in a high-velocity research environment where shipping matters as much as rigor

Responsibilities

Design and advance methods for synthesizing realistic agents from structured and unstructured profile data
Build and evaluate simulation pipelines that produce population-level distributional accuracy across demographic, attitudinal, and behavioral dimensions
Develop techniques for grounding agent behavior in real-world data sources — surveys, behavioral traces, psychometric instruments, and domain-specific datasets
Design hybrid architectures that combine large language models with classical methods from computational social science, psychometrics, and survey methodology
Build rigorous evaluation frameworks that measure simulation quality against known ground-truth distributions (e.g., GSS, census, proprietary panel data)
Identify and close gaps between simulated and observed population behavior through iterative experimentation