Research Operations, Code

Mercor•San Francisco, CA

1d•$130,000 - $250,000•Onsite

About The Position

Silicon Valley's leading AI labs partner with Mercor to build the high-quality code and reasoning data that trains their frontier models. As a member of Research Ops on the Code team, you will own multi-million-dollar data programs end to end — and ownership here is technical as much as operational. You will: Decompose frontier model capabilities. Reason about where a lab's frontier coding model is weak, and translate that into the data that will make it stronger. Design the pipeline. Shape how data gets produced — human-expert workflows, synthetic generation, model-in-the-loop systems — not just run a fixed process. Drive execution. Turn that design into delivery through a team of expert contributors, under demanding timelines, at a consistently high quality bar. Own the customer. Build deep relationships with lab researchers and become the person they trust to tell them what data they actually need. This role sits at the intersection of operator and builder. You'll spend as much time reasoning about designing tasks which fail frontier models in fair ways, designing verifiers with parity to production codebases, and constructing self-contained environments which capture real world tasks, as you will on delivery and quality. We operate with startup intensity: occasionally responsive on weekends, always a high bar. The upside is that performance incentives, high-slope career trajectory, and meaningful equity reflects that intensity.

Requirements

Technical judgment. Coding literacy and genuine ML or Model Benchmark familiarity — enough to evaluate model outputs, reason about frontier-model capabilities and failure modes, read benchmark/eval work, and translate research goals into concrete task and data designs. You don't need to be a research scientist, but you do need to go deep on the technical substance.
Operational ownership. A track record of running complex, high-stakes projects end to end, and genuine energy for large-scale execution and gritty process optimization under pressure.
Communication & customer instinct. Strong analytical and communication skills; comfortable owning high-profile relationships with technical customers.

Nice To Haves

Pipeline building: designing data or automation pipelines; hands-on with LLMs/agents; building synthetic-data or model-in-the-loop systems.
Research fluency: connecting model/benchmark literature to what data would move a frontier model; having created a benchmark or published analysis of model behavior.
Backgrounds that often fit: ML/data/software engineers who love operating, technical PMs, research engineers, or strong generalist operators with real technical range. Backgrounds from consulting, finance, or high-growth startups can work if paired with real technical fluency.

Responsibilities

Develop a working understanding of what our customers' models can and can't do, and where the capability gaps are.
Evaluate model outputs and benchmark performance, and create targeted loss analysis to identify what data will drive improvement.
Translate research goals into concrete task designs, difficulty targets, and quality specifications — including novel frontier tasks that don't exist anywhere else.
Design how data is produced — human-expert, synthetic, and hybrid model-in-the-loop pipelines — and continuously improve them.
Prototype new generation and validation approaches; bring ideas for new data products, not just improvements to existing ones.
Balance quality, throughput, and cost as you scale a pipeline from prototype to production.
Manage end-to-end data pipelines from customer specification to final delivery.
Diagnose bottlenecks, restructure workflows, and implement solutions — incentive systems, workflow re-sequencing, sharper instructions, scaled review processes, and automated quality assurance.
Run daily internal syncs ("war rooms") to stay ahead of issues.
Act as the primary point of contact for leading AI labs; deliver clear, consistent reporting.
Proactively anticipate researcher needs and identify opportunities for expansion.
Design the human-expert labeling and evaluation flows on the platform — how tasks are structured, reviewed, and scored.
Source, vet, train, and performance-manage teams of domain experts (software engineers, competitive programmers, and specialists).
Maintain a high execution and quality standard across every stage of production.