Computational Science Biology Expert for AI Benchmarking

Toptal; India; Remote; INR 1,653,000 - 3,582,000; Full time

Benefits offered by this job

Remote work

Per-task pay

Job summary

Toptal is seeking researchers to design challenging, terminal-based analysis tasks for AI agents. You will create precise problem statements, reference solutions in Python/R/shell, and automated verifications grounded in real biology workflows.

This is a remote, task-based contract paid per task. You will choose how many tasks to take on, with an emphasis on reproducible, Docker/Git‑driven packaging and authentic practice using tools like Ensembl VEP and Bioconductor. Immediate start preferred.

Qualifications

• PhD or 3+ years of professional biology research with publications or equivalent output.
• Hands‑on analytical work in the last 3 years naming tools, datasets, and rules used.
• Ability to run analyses in code (Python, R, C++) or domain software.
• Clear sense of checkable results and how to catch errors.

Responsibilities

• Design expert‑level tasks based on real analyses such as variant annotation and differential expression.
• Specify task inputs, data, software environment, and deliverables precisely.
• Write a complete reference solution in Python, R, or shell (60–75 minutes).
• Build automated verification to confirm correct outputs and document common mistakes.
• Package tasks reproducibly using Docker and Git for clean runs.
• Refine tasks based on reviewer feedback to keep them realistic and hard.

Skills

Python

R

Command line

Biology data analysis

Education

PhD in Biology

3+ years of research

Tools

Docker

Git

Ensembl VEP

Bioconductor packages

Job description

A fast-growing AI company is building expert-level benchmarks that test whether AI agents can complete real scientific workflows from start to finish, with no human help. They need working biologists to supply the hard part: realistic analyses that only someone who has done the work would know how to set up, solve, and verify.

You will turn analyses from your own research or professional work into challenging, terminal-based tasks. Each task comes with a working reference solution and an automated check that proves whether the agent got the right answer. If you have ever thought “an AI could never do this part of my job,” this project is about testing that claim properly.

This is a remote, task-based contract. You are paid per approved task, and you choose how many tasks you take on.

What You’ll Do

• Design expert-level tasks based on real analyses you have performed, such as variant annotation, differential expression, phylogenetics, structural analysis, or pipeline debugging, that an AI agent must complete end to end in a terminal.
• Specify each task precisely: the problem statement, input files and data, software environment, and the exact deliverable the agent must produce.
• Write a complete reference solution in Python, R, or shell that shows the task can be solved in roughly 60–75 minutes of focused work.
• Build automated verification, such as an output matching a reference variant catalogue or a value falling within tolerance, and document how common mistakes would be caught.
• Package tasks reproducibly using Docker and Git so they run cleanly and consistently from the command line.
• Refine tasks based on reviewer feedback, tuning difficulty, clarity, and correctness so each one is realistic, unambiguous, and genuinely hard for today’s AI agents.
• Ground every task in authentic practice, using named tools, datasets, and domain conventions such as Ensembl VEP, Bioconductor packages, or HPC workflows.

What You Bring

• PhD (completed or advanced candidacy) or 3+ years of professional research in a biology subfield, with publications, a thesis, or tangible lab or computational output.
• Hands‑on analytical work within the last 3 years, and the ability to name the specific tools, datasets, and rules you used.
• You run analyses yourself in code (Python, R, C++) or directly in domain software, rather than only managing, teaching, or reviewing them.
• A clear sense of what a checkable result looks like in your field and how you would catch an error.
• Comfort with Python (or a similar language), the command line, and working with files, CSV/JSON, APIs, or databases.
• Availability to start immediately (from early November 2026) and work on a per-task basis.

Nice to Have

• First-author papers with released code or data
• Contributions to scientific or open-source bioinformatics software
• Experience with Docker, Git, and writing automated tests
• Teaching or peer-review experience in your subfield
• Experience running workloads on HPC clusters