Aaru raised a Series A above $50 million at a $1 billion valuation before cracking $10 million in annual recurring revenue, TechCrunch reported. That gap dictates who gets hired: the company pays $200,000–$600,000 for individual contributors who can both publish at NeurIPS and ship Kubernetes-native agent runtimes to production. The hiring bar, proven research impact and software craftsmanship in the same person, shapes a workforce built for cross-disciplinary execution across New York and Singapore.
The company organizes around three interconnected tracks: simulation engineering, prediction research, and product delivery. Zero G Talent's board shows six active roles, all based in New York: Head of Simulation Engineering, Evaluation Researcher, Prediction Researcher, Product Engineer, Platform Engineer, and Data Engineer. Salary bands cluster between $200,000 and $600,000 for individual contributors, with the simulation-engineering lead ranging to $750,000, Zero G Talent's data shows, a spread reflecting the premium on people who move fluidly between model architecture and distributed systems. The board's median across 18 salaried listings sits at $425,000.
| Role | Salary Band (USD/year) | Location |
|---|---|---|
| Head of Simulation Engineering | $450,000–$750,000 | NYC |
| Evaluation Researcher | $200,000–$600,000 | NYC |
| Prediction Researcher | $200,000–$600,000 | NYC |
| Product Engineer | $200,000–$600,000 | NYC |
| Platform Engineer | $200,000–$600,000 | NYC |
| Data Engineer | $200,000–$600,000 | NYC |
Simulation engineering owns the agent runtime. The Head of Simulation Engineering role calls for someone who has built and operated large-scale multi-agent systems, not just prototyped them. That means Kubernetes-native serving stacks, custom scheduling for heterogeneous workloads, and observability that catches distributional drift before it contaminates a client's go-to-market decision. John Kessler, co-founder and CTO, set the technical bar during his time at MIT's City Science Lab, where he built urban behavior simulation, the closest academic analog to what Aaru now runs in production. He still reviews every simulation-engineering candidate's system-design write-up before a final round.
Research splits into two lanes. Evaluation Researchers design the validation frameworks that tell a client whether a synthetic population actually mirrors its real-world counterpart. The job requires fluency in causal identification, survey methodology, and the statistical quirks of LLM-generated outputs. Prediction Researchers own the forecasting models that turn simulated behavior into probability distributions over business outcomes: demand curves, vote shares, churn rates.
Product and platform engineers close the loop. Product Engineers embed with clients (Accenture, EY, Interpublic Group, political campaigns) to translate vague strategic questions into simulation specifications. Platform Engineers build the internal tooling that lets researchers spin up a 50,000-agent cohort in minutes rather than days. Data Engineers maintain the ingestion pipelines that fuse financial transactions, public records, media consumption, and proprietary panels into the feature store every model draws on. All three engineering tracks share the same $200,000–$600,000 band, signaling that Aaru treats infrastructure craft as equally critical to research novelty.
The "about" page adds "strategic finance" to the experience mix, hinting at a small but deliberate operations layer that handles the economics of running inference at population scale. Headcount grew from 21 employees in March 2023 to 27 by April 2026, with a stated goal to more than double in 2026. The current team is roughly half engineering, a quarter research, and the remainder split across product, finance, and go-to-market.
Equity details do not appear in Aaru's public disclosures or the board postings. The funding history shows a multi-tier valuation structure, some equity priced at $1 billion, other tranches lower, producing a blended valuation below the headline number. TechCrunch reports this mechanism is becoming more common for desirable AI startups. Early employees' strike prices and percentage grants will depend on which tranche their grant pulls from and when they join relative to the next 409A refresh. Candidates should ask directly about the current 409A, the preferred-price-to-common discount, and whether refresh grants follow a fixed schedule or performance triggers.
Benefits are not itemized in the board data or the company's career materials.
The compensation picture that emerges is deliberate compression at the IC level with a sharp step up for simulation leadership, cash-heavy bands that compete with Manhattan's top AI labs, and equity terms that will only clarify at offer stage. For a candidate, the leverage point is research impact: the $600,000 ceiling on the research and engineering tracks is not a theoretical maximum; it is the number Aaru puts on a proven record of moving simulation accuracy or shipping platform capabilities that directly improve prediction correlation. The median of $425,000 across 18 roles suggests most hires land in the upper half of the band, not the middle.
How the Loop Runs
Aaru's public recruiting footprint is thin. The company does not publish a hiring handbook, and employee accounts of the interview loop are scarce in open sources. What exists are the job postings themselves, six roles listed on Zero G Talent as of late 2025, and a growth trajectory that implies a process moving faster than most.
The roles reveal the screening bar. "Head of Simulation Engineering" at a $450,000 floor signals ownership of the agent architecture end-to-end: memory systems, retrieval, multi-agent orchestration, evaluation harnesses. The three researcher titles, Evaluation, Prediction, and the implied core simulation work, point to a hiring priority on measurement methodology. Aaru's product lives or dies on correlation metrics; the company cites 0.90 median correlation on EY's wealth research, replicated in a day versus six months. Candidates who have designed evaluation frameworks for generative systems, not just trained models, match the brief. Product and Platform Engineers sit in the $200,000–$600,000 band typical for senior ICs at well-capitalized AI startups; the breadth suggests they need engineers who can move across the stack from agent runtimes to customer-facing APIs without handoff friction.
The company's "about" page frames the mission as "grounding large-scale populations of AI agents in real demographic, behavioral, and outcomes data." Candidates whose experience is purely LLM prompting or synthetic data generation without validation against ground-truth behavioral data will not map to the problem set. The founders, Cameron Fink, Ned Koh, and John Kessler, built the simulation engine themselves before hiring.
What the research does not show: number of stages, take-home assignments, live coding format, panel composition, timeline from application to offer, or any documented "culture add" rubric. No Glassdoor-style accounts from candidates exist in the sources. The board data confirms the roles and bands; the rest is inference from the technical surface area the roles imply. If you apply, prepare to discuss how you would evaluate agent populations against real-world outcomes, not just perplexity or benchmark scores. That is the problem Aaru pays to solve.
Where You'll Sit
Aaru operates from two hubs: a six-story former architect's office in Tribeca, New York, and a presence in Singapore noted on the company's about page. As of April 2026, the Tribeca space houses the full 27-person team; every role listed on the Zero G Talent board is tagged NYC. The lease is short-term by design. The founders have already selected the new space and plan to move within weeks, targeting a headcount that more than doubles the current roster within the year.
Aaru's product is a simulation engine that spins up thousands of AI agents grounded in demographic, behavioral, and outcomes data. The compute stack comprises GPU clusters for training and inference, data pipelines ingesting thousands of sources from financial transactions to media consumption, and the infrastructure to serve near-instant predictions to customers.
Singapore appears in the company's self-description ("We are based in New York and Singapore") but the research does not specify its function: whether it hosts a satellite engineering team, a sales and partnerships office, or a research outpost. No roles on the job board carry a Singapore tag as of the latest postings. For candidates, this means the actionable location decision is binary: New York today, New York (SoHo) tomorrow, with Singapore as a future variable rather than a current option.
The move to SoHo signals more than square footage. The founders cite "Win together: Winning compounds. Individual leverage grows when knowledge and trust are shared across the team" as a cultural pillar. The Tribeca space, a repurposed architect's office, already carried that DNA, a layout that forces collision. The new loft doubles down on it.
For an engineer or researcher evaluating the offer, the physical reality is straightforward: you'll sit in Lower Manhattan, first in Tribeca, then in SoHo, surrounded by the simulation, platform, and data teams that ship the product. Singapore exists on the map but not on the org chart yet. If your work demands a GPU cluster, a clean data lake, and a team that treats simulation fidelity as a craft problem, the loft is where that work happens.
Who Lasts
The people who stay at Aaru tend to share a specific kind of impatience with the gap between what people say and what they do. That gap is the company's entire premise: stated intentions correlate poorly with actual behavior, and the team's job is to close that distance with simulation. Engineers and researchers who find that problem intellectually unavoidable — not just commercially interesting — are the ones who last.
The signal shows up early in the founders' own trajectory. Fink and Koh met in high school bonding over hacking projects and boredom with coursework; both dropped out of elite colleges within weeks. Kessler brought simulation research from MIT's City Science Lab. That pattern, self-directed builders who left conventional credentialing paths because the work mattered more, recurs in the hiring data. The compensation bands reflect a bar that selects for that same combination over pedigree alone.
Ethical clarity is another filter. The founders describe "very strict ethical rules": no work for actors who promote or create violence, a nonpartisan stance across political engagements, and an explicit rejection of the Cambridge Analytica model. Koh framed it as "almost a moral mission", giving decision-makers numbers accurate enough to take real risks. Employees who need clear ethical guardrails to do their best work find them here; those who want to push boundaries without constraint do not.
The work itself demands cross-disciplinary fluency. Aaru's simulations pull from thousands of data sources, those same sources, to spin up massive custom populations. The use cases span product innovation, marketing, audience segmentation, scenario planning, and strategic communications. A 0.90 Spearman correlation against EY's unreleased Global Wealth Survey — achieved in a day rather than six months — is the kind of validation that matters to people who care whether their models actually predict.
Speed compounds the difficulty. The company moved from a six-story Tribeca office to a SoHo artist's loft within months, planning to more than double headcount again this year. The founders talk in 80-year horizons — "I'll be working here for the next 80 years," Fink said — while shipping at high velocity. That tension, long-term conviction paired with extreme short-term velocity, selects for people who can hold both without burning out.
External scrutiny adds pressure. A New York Times op-ed labeled the approach "silicon sampling" and warned of a broken information ecosystem. Competitors like Simile have avoided political work entirely. Aaru has not: it called the New York Democratic primary and the mayoral race, and plans to engage in midterm campaigns "that inspire it." Working on nuclear power, reproductive rights, and school choice narratives requires a temperament that treats controversy as signal, not noise.
The clearer signal is who self-selects in. The roles on the board, simulation engineering, evaluation research, prediction research, platform, product, data, all sit at the intersection of ML systems and human behavior modeling. Candidates who have published in both venues, or built systems that failed in production because they ignored the behavioral layer, tend to recognize the problem immediately. They don't need to be sold on the mission. They've already been looking for it, the same way Kessler did at MIT, the same way Fink and Koh did in a dorm room, the same way the next hire will when they walk into the SoHo loft and see the simulation running.
Working in frontier tech? Zero G Talent tracks the openings: see every open Aaru role, browse frontier tech jobs, the companies hiring, and the people building the field.