Research Engineer

Description

The company has built AI-moderated user interviews, a new way to run research.

Global brands including HubSpot, Microsoft, Uber and Nestle use it to get deep qualitative insight into their users, quickly and at scale.

The research market is worth over $140B and still growing. Incumbents are slow, outdated and expensive to use, which leaves room for the company to take share.

Top Silicon Valley investors back the company. It closed a $30M Series B in December 2025, less than six months after its Series A. Radical Ventures led the round, with participation from M12 (Microsoft's venture arm), 8VC, Y Combinator and Adverb Ventures.

The company's business grew 8x in its breakout year as enterprises across industries adopted AI-moderated research.

The company is a close-knit team serving some of the largest enterprises in the world.

The Digital Twins Team builds AI systems that represent specific real people and answer research questions the way those people would. It runs an ongoing program measuring how closely these systems reproduce human responses and behavior across several levels of analysis.

This role reports to the Head Scientist. The ideal candidate has supported scientists in an academic computational social science lab or similar research setting and has hands-on experience building systems on large language models.

Responsibilities

• Build and run the evaluation infrastructure that compares modeling approaches, so new tests can be set up quickly, run in parallel and automatically, and monitored.
• Turn the scientists' specifications into measurement pipelines with deterministic, traceable outputs.
• Version evaluation datasets and keep held-out sets out of training.
• Create and maintain pipelines for human-participant data, external datasets and model training data, including linking records that belong to the same participant across sources.
• Keep research code aligned with the production product so validated improvements ship quickly.
• Find and prevent anything that threatens the validity of evaluations.
• Track compute and API spend across experiments and what it implies for production costs.
• Manage repositories, code review, documentation and the reproducibility of the team's coding workflow.
• Spot implementation choices that could be more correct or efficient, and propose changes.

Requirements

• Several years of software engineering experience, including supporting researchers in a lab or similar setting.
• Ability to turn scientific requirements and experiment designs into working code.
• Experience building and running large-scale pipelines on human behavior data, and confirming they match the criteria and design set by the researchers.
• Strong Python, with the skill to read and audit code produced by agentic coding tools.
• Solid grasp of computing fundamentals: file systems, software libraries, databases, SQL, tabular data and Git-based workflows.
• Experience with cloud infrastructure and running many jobs concurrently.
• Working knowledge of machine learning, covering how transformer language models behave, how model APIs are used and what fine-tuning involves.
• A habit of writing code that is deterministic, reproducible, cost-aware and well documented.
• Preferred: hands-on experience building LLM systems such as prompt pipelines, evaluation harnesses, embedding computation, agentic systems and/or fine-tuning infrastructure.
• Preferred: fine-tuning open-weight language models on multi-GPU hardware.
• Preferred: serving open-weight models with an inference framework such as vLLM or SGLang, including batched inference and prefix caching for many requests sharing a common prompt.
• Preferred: experiment tracking and cost monitoring for machine learning workloads, including GPU provisioning across cloud providers.
• Preferred: work with survey or panel data and data collection platforms.
• Preferred: extracting and storing model activations for downstream analysis.
• Preferred: prior work on generative agents, digital twins or simulation of human respondents.