The right minds for
harder AI problems.
We connect AI research labs and AI data companies with technical contributors evaluated on the work itself: coding, reinforcement-learning environments, agent evaluations, verifiers and AI training tasks.
Blackdrome builds benchmarks, environments and verifiers. We assess contributors with the same standards we use for our own research.
Alreadydoingthiswork,atscale.
Engineers, researchers and reviewers across coding, RL environments, agent evaluation and technical review.
Experts who have done this work at top tech companies and AI labs, vetted before they join.
Our team has run expert delivery for AI training and evaluation programs. We know what a good brief, a reliable contributor and an on-time delivery take.
Frontier AI labs
Post-training and evaluation teams that need specialists who can write hard tasks, judge model output and find where a model breaks.
AI data companies
Delivery teams that need dependable capacity for coding, RL and evaluation projects, with quality that holds as the project scales.
Research teams
Academic and independent groups building benchmarks, environments and datasets that need contributors who understand the research.
Fourareas,chosenonpurpose.
Contributors who work in real codebases.
Repository-level tasks and debugging, test writing, code review and reference solutions, across Python, TypeScript, Go, Rust, C++ and more.
People who build the places where models learn.
Environment and task authoring, verifier and grader design, reward-hacking and shortcut checks, and difficulty calibrated against current models.
Reviewers who can say exactly what went wrong.
Trajectory review and scoring, rubric design, failure analysis and red teaming, for multi-step and tool-use tasks.
Senior reviewers who keep a project consistent.
Expert QA of tasks and labels, adjudication of disagreements, guideline audits, and feedback that raises a team's bar.
Evaluatedonthework,beforeyoumeetthem.
Skill assessment
A work sample in the contributor's own area, the same kind of task they would do for you, scored against a written rubric. Not a quiz or a keyword screen.
Technical review
A reviewer from our team reads the work and the reasoning behind it: correctness, whether they verify their own answers, and shortcuts that would pass a weak check.
Availability verification
Hours, time zone, start date and tooling are confirmed before anyone is proposed for a project, so the people you meet can actually start.
Prefertotalkfirst?
Write to us about a project, a partnership or joining the network. Conversations about your work stay confidential.