The work
You will turn real Earth science workflows into self-contained terminal tasks for testing AI agents. The work covers environmental, climate, atmospheric, geospatial, geological, hydrological, oceanographic, and time-series data.
You will build reproducible computing environments, prepare expert solutions, and set clear standards for grading results. You will also check scientific assumptions, data quality, accuracy, and the limits of each task.
- Create benchmark tasks based on authentic Earth science workflows.
- Prepare and process climate, atmospheric, geological, hydrological, oceanographic, and geospatial datasets.
- Build reproducible environments with scientific libraries and command-line tools.
- Write solutions in Python, R, Bash, Julia, or a comparable scientific language.
- Develop tasks involving numerical modeling, interpolation, forecasting, remote sensing, and environmental risk analysis.
- Define grading criteria for data changes, model behavior, and spatial or time-based accuracy.
- Validate coordinate systems, units, timestamps, missing data, file formats, metadata, and scientific assumptions.
- Create automated tests and document data sources, expected outputs, edge cases, and limitations.
What it pays and takes
This is a remote contractor assignment scheduled for five weeks. The work requires strong Earth science knowledge, scientific programming ability, and experience with reproducible technical workflows.
- $300 per approved task.
- At least 20 hours per week, including 4 hours of overlap with Pacific Time.
- Open worldwide.
- Professional fluency in English.
- A Ph.D., postdoctoral experience, or equivalent advanced technical experience in Earth sciences or a closely related field.
- Strong knowledge of climate science, atmospheric science, geophysics, oceanography, geology, hydrology, remote sensing, environmental modeling, or Earth-system science.
- Strong programming ability in Python, R, Julia, Bash, or a comparable scientific language.
- Hands-on experience with scientific data processing, numerical modeling, geospatial analysis, environmental datasets, or time-series analysis.
- Ability to work independently in Linux and terminal-based environments.
- Strong understanding of scientific quality control, uncertainty, numerical accuracy, and reproducible workflows.
- Useful experience includes NumPy, pandas, SciPy, xarray, rasterio, GeoPandas, Cartopy, GDAL, NetCDF, HDF5, GeoTIFF, shapefiles, or GRIB.
- Additional useful experience includes Docker, Conda, Git, CI/CD, automated testing, high-performance computing, scientific benchmarking, automated graders, or evaluating AI coding agents.
- Publications or open-source work in Earth, environmental, geospatial, or computational sciences is valuable.
How it works
Apply on OpenTrain with your resume, then complete the application on the hiring site. OpenTrain is a platform where people find AI training and data-labeling work, build a profile, and start applications.
About AI training work
AI training is the human work behind systems that learn from examples, including preparing data, testing model behavior, and reviewing results. People with specialized scientific knowledge help create accurate examples and evaluations for technical AI systems.