Founding Data Lead in London

Full-time · London or San Francisco Bay Area, hybrid.

About us

Science runs on trust, but the machinery meant to produce that trust is failing.

Peer review is slow, opaque, inconsistent, and increasingly gamed.

And a flood of AI-generated research is about to overwhelm a system that was already straining.

No one is the credible, independent arbiter of research quality.

We think there should be one.

XSci is building a certification layer for science: expert peer review and AI working together, so researchers, institutions, and AI developers can assess research quality quickly and at scale.

Every expert review on XSci also produces something rare: a senior scientist's judgment on frontier work, including where that judgment overrules the AI.

Capturing that record properly is half of what we are building.

The role

You will own XSci's data: how it is captured, how its quality is proven, and who may use it.

It is the human data problem that frontier AI runs on, applied to a domain where the annotators are Ph Ds and the labels are scientific arguments.

This is a founding role.

You report to the founder and work daily with our Chief Scientist and CTO.

There is no data team yet; you are the first person in it, and you will decide what it becomes.

What you'll do

• Design the data model for expert evaluation: task schemas, rubrics, gold sets, inter-annotator agreement, reviewer quality scoring.
• Capture the reviewing process itself, not just its output: draft history, revisions, and the reasoning trail behind a finished review.

Instrument it end to end and turn it into something analysable.

• Build provenance, consent, and usage rights into every record from the start, so we always know where a record came from and what may be done with it.
• Run expert labeling programs end to end, sourcing specs, pilot batches, calibration, throughput and cost tracking, and the loop back into task design.
• Stand up the eval harness that benchmarks AI review output against expert ground truth, and keep it honest as the models change.
• Make the dataset legible to a demanding outside reader: sample packs, data cards, delivery formats, and a view on what the next batch should contain.

You are the technical counterpart when someone outside the company asks how this was made.

• Work with the founder on how the data is governed and who gets access to it, and be the person who can explain it to a technical audience outside the company, including in diligence.
• Ship the production data services (orchestration, storage, observability) that the model and product teams depend on daily.

What we're looking for

• You have owned, not maintained, the systems behind an expert data program: specialists doing structured work against a rubric, at a quality bar somebody was paying for.

That might have been at one of the companies building human data pipelines for frontier AI, inside a lab's own human data or RLHF team, or somewhere that uses none of those words.

The title varies (data engineering, applied AI, technical program management, data operations) and the experience is what counts.

• You have been on at least one side of a frontier-lab data purchase, as the vendor delivering it or the lab receiving it, and you know what the other side looks at.
• Strong Python and SQL. Comfortable with distributed processing, workflow orchestration, and cloud infrastructure.
• Data quality as a discipline: you have debugged a bad gold set, caught a drifting annotator cohort, and rewritten a rubric that was quietly ambiguous.
• Able to work directly with domain experts and turn fuzzy scientific judgment into measurable labels.
• Senior enough to carry this alone from day one. We care less about years than about having run the whole loop, from task design to a dataset someone outside your team accepted.
• Nice to have
• A Ph D, or time spent in or around academia, scholarly publishing, or scientific tooling.

If you have written and refereed papers yourself, you will understand our annotators quickly.

• LLM evaluation, LLM-as-judge pipelines, or preference data collection.
• Why this role
• Peer review is one of the few remaining bottlenecks on the rate of science, and nobody has built the data infrastructure for it yet.

You would be defining what "good review" means in machine-readable terms, for a system that researchers actually use.

Compensation and how to apply

• Salary plus meaningful equity; this is a founding position and the equity reflects that.
• Location: London or San Francisco Bay Area, hybrid.
• Visa: we can sponsor work visas for the right candidate.
• #J-18808-Ljbffr