Applied Research Engineer

Primis; New York (NY); On-site; USD 300,000 - 500,000; Full time

Benefits offered by this job

Top-of-market compensation

Flexible base/equity packages

On-site work at NYC

Job summary

Primis in New York City is seeking a hands-on engineer to build the next generation of simulation platforms. You will design and run multi-agent systems, orchestrating thousands of LLM agents, each with memory and state, to test ideas against audiences, culture, brands, and real behavior.

This intense on-site role rewards experimentation, calibration, and rapid learning, with a focus on translating research into production-ready pipelines and scalable serving stacks.

Qualifications

• Hands-on experience in simulation, synthetic populations, forecasting, causal inference, or behavioral data.
• Experience building LLM multi-agent systems in production or research.
• Experience deploying open-weight models and serving stacks; profiling bottlenecks.

Responsibilities

• Design and build multi-agent simulation systems with thousands of LLM agents.
• Model interactions of ideas with audiences, culture, brands, and market signals.
• Calibrate prompts, bias, variance, and drift against real human responses.
• Build experimentation, calibration, uncertainty, and validation pipelines.
• Serve open-weight models on in-house clusters at scale with fidelity.

Skills

Simulation experience

ML engineering

Open-weight models deployment

Tools

KV/prefix caching

Quantisation (AWQ, GPTQ, FP8)

Multi-LoRA serving

Job description

New York City (on-site) | $300k-$500k base + significant founding equity

The role

The lab's creative models surface exceptional concepts. You'll build what comes next: the system that tests those ideas against the world they're meant to enter, including audiences, culture, brands and real behaviour. You'll then turn the results into feedback the models can learn from. The aim isn't to reward safe ideas. It's to find out whether something extraordinary can actually land.

What you'll do

• Design and build multi-agent simulation systems that orchestrate thousands of LLM agents, each with its own profile, memory and state
• Model how ideas interact with audiences, culture, brands and market signals
• Tackle the hard problems of synthetic audiences: positivity bias, low variance between agents, prompt sensitivity and drift into easy consensus. You'll calibrate everything against real human responses
• Build experimentation, calibration, uncertainty and validation pipelines tied to real-world outcomes
• Serve open-weight models on in-house clusters at scale, without changing simulation fidelity

What we're looking for

• Hands-on experience in at least one of: simulation, synthetic populations, forecasting, causal inference, recommendation/ranking, survey or behavioral data, or computational social science
• You've built LLM multi-agent systems in production or research (roles, memory, message passing, turn-taking)
• You've deployed open-weight models and know the serving stack well: KV/prefix caching, quantisation (AWQ, GPTQ, FP8) and multi-LoRA serving. You can profile and fix a bottleneck
• Strong ML engineering, software fundamentals and statistical instincts
• A research mindset: you can turn an ambiguous question into a testable system and know when the evidence is weak

Strong plus:

founding or early-stage experience, a track record of taking research through to product, and taste or fluency in a creative field such as advertising, film, design, writing, music, art or games.

• A founding seat on a problem almost nobody else is tackling, with real influence over technical direction and vision
• Top-of-market compensation, with flexible base/equity packages
• A fast-moving process with a high-calibre team

Worth knowing up front:

this is an intense, in-person environment, with at least 4 days on-site and a 6-day week. It's built for people who want to go all-in on something ambitious.