The work
You will start with a recent computational biology paper and its released code, ideally from your own research. Day to day, this means removing the implementation of the paper's core contribution, packaging the remaining repository as an agent-solvable task, writing a precise task statement, and preparing reproducible benchmarks. You will also create a hidden reference solution and a second implementation that demonstrates the stated performance bar can be exceeded.
Evaluation design and review
Each task needs held-out workloads, per-metric pass thresholds, and safeguards against benchmark leakage or superficial solutions. After agents attempt the task, you will inspect their code and results, determine whether failures come from the implementation or the evaluation harness, and assess whether alternative approaches satisfy the intended research objective.
What the platform screens for
The platform looks for a PhD in computer science or a related field, completed or in progress, along with published computational biology research and released code for at least one paper. Screening is likely to examine concrete publication and artifact details, your command of benchmarking choices, and your ability to explain measured improvements under technical follow-up. First-author work, Docker experience, and depth in read alignment, genome assembly, indexing, or variant calling are preferred, and the application process includes background checks.
Logistics
The engagement is fully remote, asynchronous, and flexible. Each task is described as taking a few hours, with later review work when agents submit attempts.
Pay band: $80/hr