Scientific Computing / Research Engineering Expert: Earth Sciences

What you'll do

This is task authoring for an AI benchmark, Terminal Bench Science, in the Earth-science corner. You take a workflow you would run for real (processing a climate or ocean time series, reprojecting geospatial data, running an atmospheric model) and turn it into a self-contained terminal task an AI agent has to solve, with a reproducible environment and an objective way to grade the result.

Concretely the ad lists:

• Preparing geospatial, climate, atmospheric, geological, hydrological or oceanographic datasets
• Building reproducible environments with the right scientific libraries and command-line tools
• Writing expert solutions in Python, R, Bash, Julia or domain software
• Defining grading criteria and automated tests for numerical tolerances, file formats, metadata and reproducibility
• Checking coordinate systems, units, timestamps and missing-data handling
• Debugging projections, large datasets, dependencies and numerical stability, then documenting data provenance, edge cases and limits

Who fits

• PhD, postdoctoral experience or equivalent advanced experience in Earth sciences or a close field
• Depth in at least one area: climate, atmospheric science, geophysics, oceanography, geology, hydrology, remote sensing, environmental modelling or Earth-system science
• Strong programming in Python, R, Julia, Bash or similar, comfortable in Linux terminals
• Hands-on scientific data processing, modelling, geospatial or time-series work

Preferred: NumPy, pandas, SciPy, xarray, rasterio, GeoPandas, Cartopy or GDAL; formats such as NetCDF, HDF5, GeoTIFF, shapefiles or GRIB; Docker, Conda, Git, CI or HPC familiarity. Nice to have: benchmark or grader development, evaluating AI coding agents, publications or open-source contributions.

What it pays

$300 per approved task, as the ad states. Only tasks that pass review count, so earnings depend on your approval rate and how fast you build a task. The ad does not say how many tasks to expect over the contract.

Worth knowing

Good:

• A published, per-task rate
• Work that uses real domain and coding skill rather than generic labelling
• Fully remote

Less good:

• Pay is per approved task, so rejected or reworked tasks cost you time
• The commitment is heavy: at least 6 hours a day and a minimum of 40 hours a week, with 4 hours of overlap with Pacific Time
• Short engagement: 5 weeks, with the start date given as the following week
• Contractor assignment with no medical or paid leave
• The ad does not describe an assessment; Turing's help centre describes skill tests and, for some roles, a live interview

About this listing

Posted by Turing on 21 September 2026 and read on its jobs board on 18 October 2026. Rate, hours and contract terms are Turing's own text; the commentary is ours. See Turing.