Research Partnerships Lead: Model Behavior, Evals & Interpretability | San Francisco (On-site) | Up to $300K Base, $500K OTE + Equity — San Francisco, CA

"One accurate measurement is worth a thousand expert opinions." Grace Hopper

Grace Hopper had a point. But what if you're the rare person who brings both: the measurements and the opinions?

That's exactly who this role is for. Our client is one of the fastest-growing AI data companies in the world, a unicorn building the datasets and evaluations that sit behind today's most capable models. Nearly every frontier lab is a customer. The team is still under 100 people, the founders came out of elite quant trading and big tech, and the growth curve points almost straight up.

They're hiring two Research Partnerships Leads in San Francisco, and it's one of the more unusual roles we've worked on this year.

THE ROLE IN ONE LINE

You're the researcher on the commercial team.

The people who buy frontier data are researchers inside AI labs. They have no patience for a pitch; they want a peer. They want someone who has stared at model outputs long enough to know why an assistant feels sharp for one user and useless for another, and who can back that view with evidence.

Your job is to hold those views, sharpen them, and turn them into proposals, analyses and research collateral that land on a lab researcher's desk and earn the reaction every vendor wants: "these people actually get it."

You'll start behind the scenes, writing and building alongside the commercial team. Over time you'll step in front of lab clients yourself. With only two hires to begin with, you won't inherit a function. You'll shape one.

WHAT YOU'LL ACTUALLY DO

• Decide what good looks like. Form clear, defensible views on how a personal assistant, an enterprise agent or a general chatbot should behave for different kinds of users.

• Write things researchers finish reading. Proposals and analyses that show where models fall short and which data would close the gap.

• Build the evidence. Run the experiments and pull the examples that turn an opinion into an argument.

• Present it live. First to the internal team, then to rooms full of lab researchers who will ask hard questions.

• Listen closely. Hear first-hand how different labs think about model behavior, evals and data, and feed that back into what gets built next.

THE SCORECARD: HOW YOU'LL BE ASSESSED

Most adverts make you guess what the hiring team cares about. We'd rather show you. Every candidate is measured against the same scorecard, and here it is in plain English.

The four essentials

• You've worked where the frontier is. Time at a frontier or neo lab, or at an AI evals or data company. Think OpenAI, Anthropic, Scale AI, Snorkel AI, METR or Redwood Research, and their peers.

• Your model behavior or interpretability work was hands-on. Post-training, evals, answer quality, interpretability methods: you read the outputs, designed the experiments and drew the conclusions yourself.

• You write and present with real clarity. Could you write a proposal that a lab researcher would find sharp and credible, then present it live to a room? Both halves count.

• You can be in the office five days a week. Monday to Friday in San Francisco's Financial District. Relocation support is available.

The baseline

• Roughly a year or more of model behavior, interpretability or evals work. Quality beats tenure. Eight strong months on model behavior inside a frontier lab counts for more than several years somewhere adjacent.

• Strong, specific opinions. "It depends" won't carry you far. You can say what makes a model good or bad for a particular user or use case, explain why, and defend it.

What makes you stand out

• Contributions to benchmarks, datasets or evaluation projects, or work with human annotator and expert pools.

• A technical degree in computer science, machine learning, statistics, data analytics, economics or similar, or equivalent hands-on experience.

Where the fit gets harder

• Product-facing roles that never involved hands-on model work.

• AI experience gained solely at a large cloud provider, with no frontier or neo lab chapter alongside it.

WHY THIS ONE IS WORTH YOUR ATTENTION

• Reach. Inside a single lab, your thinking influences one family of models. Here it travels across almost the entire frontier at once.

• A front-row seat. Learn how researchers at different labs really think about behavior, evals and data.

• Serious backing. One of the deepest ready-made data catalogs in the business behind every proposal you write.

• A blank page. Two founding hires, a direct line to the leader building the function, and room to grow it.

• Proper reward. Up to $300K base, on-target earnings up to $500K, plus competitive equity.

THE DETAILS

• Location: San Francisco, Financial District

• Work pattern: On-site, Monday to Friday

• Base salary: Up to $300K

• On-target earnings: Up to $500K

• Equity: Competitive

• Relocation: Support available

• Work authorization: Existing US work authorization strongly preferred

7XT 32tR 9pnZgZ3: YOUR TRUE PARTNERS IN AI AND TECH ROLES

aqh FqcG YH8awbc is a specialist recruitment firm focused on AI and technology. We keep our desk deliberately small so that every candidate deals with a person who understands the work, knows the client and has read the scorecard before reading your resume. We prepare you properly, we tell you where you stand, and we stay in your corner from first conversation to signed offer.

READY?

Apply here on LinkedIn and speak directly to the scorecard. Tell us which lab or evals company you worked at, what you personally did on model behavior, and the one view on model behavior you'd defend in a room of researchers.