The work
You will start with a recent algorithms paper that has a released implementation, ideally one of your own. You will remove the code implementing its central contribution, preserve enough of the surrounding repository to make the problem solvable, and write a task that asks an AI agent to reconstruct the missing method and improve on the authors’ benchmark results.
Evaluation design
Each task needs held-out workloads, metric-specific pass thresholds, a hidden reference solution, and a second solution demonstrating that the stated bar can be exceeded. You may later review agent submissions for correctness, performance, robustness, and compliance with the task specification. Relevant research areas include graph, approximation, streaming and online algorithms with reference implementations.
What the screen assesses
The platform looks for a PhD in computer science or a related field, completed or in progress, along with published algorithms research and released code. Expect follow-up on a specific paper, your implementation contribution, the evidence behind reported improvements, and how you would prevent benchmark leakage or fragile scoring. First-author work with measured gains, Docker experience, and reproducible benchmark practice are preferred rather than stated as universal requirements.
Logistics
The contract is fully remote, asynchronous and flexible. Each task typically takes a few hours, though the scope will depend on the paper, repository and evaluation design. Pay band: $80/hr. The application process includes background checks.