Agentic Life Science Expert

What you'll do

You build test problems for AI agents, not for people. Each task is a small piece of realistic computational biology: the agent gets input files and instructions, then has to inspect the data, pick a method, run the analysis, deal with whatever goes wrong midway and hand back a scientifically valid result. The aim is to see whether a model can do real multi-step scientific work rather than answer a quiz question.

For every task you also produce the other half of the package:

• Realistic datasets, input files, constraints and expected deliverables
• A reproducible expert solution and an objectively checkable ground truth that runs in the provided sandbox
• Automated or semi-automated grading criteria
• Validation of the assumptions, calculations, code and intermediate outputs

Tasks must be self-contained and run in network-isolated environments, so nothing can depend on downloading a database at run time. You also respond to reviewer feedback while keeping throughput up.

Who fits

Required:

• A Ph.D., postdoc or equivalent research background in the life sciences with strong computational work behind it
• Strong scientific programming, particularly Python
• Experience running multi-step computational analyses
• The ability to check both the biology and the code output yourself

Preferred: bioinformatics, computational genomics, systems biology, computational neuroscience, biostatistics, computational drug discovery, computational biochemistry, structural biology, protein engineering or computational microbiology; scientific libraries and command-line tools; reproducible pipelines; and skill at turning a real research workflow into a bounded, gradable task. The ad sets no country requirement.

What it pays

The ad states $280 per approved task. Note the word approved: tasks that reviewers reject presumably do not pay, so your hourly result depends on how quickly you can produce work that passes. The ad does not say how many tasks to expect or how long one takes.

Worth knowing

Good:

• A clear per-task rate, higher than most Turing ads, which usually publish none
• Work that uses real domain and coding skill rather than generic rating
• Short, defined engagement of 5 weeks

Less good:

• Heavy time commitment: at least 6 hours a day and a minimum of 40 hours a week, with a 4-hour overlap with PST
• Pay only for approved tasks, so rejected work is unpaid effort
• Contractor assignment with no medical or paid leave
• The ad says the start is "next week", so the process is likely fast; the ad does not describe assessments

About this listing

Posted on Turing's job board on 21 September 2026 and read on 18 October 2026. The rate, hours, duration and requirements are Turing's own text. Turing's help centre describes skill tests and, for some roles, a live interview; see the company page. See Turing.