Job Category: Coding & STEM

We are seeking experienced in Biology help evaluate and enhance large language model (LLM) outputs related to nutrition, dietary habits, and healthy eating. You’ll assess AI-generated content for scientific accuracy, safety, and relevance — contributing to the development of responsible AI in nutrition and wellness. Selected talents will label and evaluate data related to food intake, dietary patterns, and nutrition information to support AI models focused on nutrition tracking, coaching, or research. Main tasks listed: Annotate food items, portion sizes, meal composition, and nutrient content from text, images, or logs; Classify dietary patterns (e.g., macronutrient balance, meal timing, dietary restrictions/preferences); Verify accuracy of nutritional data against trusted references (e.g., food composition databases); Identify and flag mislabeled, inconsistent, or culturally ambiguous food entries.

In this role, you will work on cutting-edge initiatives to fine-tune large language models by leveraging your deep analytical skills and mathematics expertise. The ideal candidate possesses a strong foundation in Mathematics —spanning advanced engineering entrance levels through graduate and PhD-level concepts—and the ability to break down complex phenomena into clear, step-by-step reasoning. Main tasks listed: In this role, you will help set the benchmark for AI capabilities in advanced physical sciences. Your day-to-day responsibilities will include:; Problem Creation & Solutioning: Designing and solving challenging physics problems that test the boundaries of large language models; Authoring Gold-Standard Data: Creating clear, high-quality, step-by-step solutions with detailed and articulated reasoning; Research Collaboration: Working with LLM researchers to align task designs with evaluation goals, targeting key areas where models struggle (e.g., symbolic manipulation, abstraction, multi-step reasoning).

Commitments Required: At least 6 hours per day and a minimum of 40 hours per week, with a 4-hour overlap with PST. Employment Type: Contractor assignment; no medical or paid leave. Main tasks listed: Design challenging, realistic agentic scientific workflows across the life sciences; Create tasks requiring multiple computational steps rather than single-script or simple question-and-answer solutions; Develop realistic input files, scientific datasets, instructions, constraints, and expected deliverables; Build tasks requiring agents to inspect data, select appropriate methods, execute analyses, troubleshoot problems, and synthesize results.

We are seeking Bioinformatics and Computational Genomics experts to design challenging, agentic genomics tasks and reference solutions for GeneBench Pro. You will create realistic computational biology problems that require an AI agent to interpret biological data, write and execute code, work across scientific files, and produce objectively verifiable outputs. Main tasks listed: • Design novel, challenging tasks in computational genomics and bioinformatics; • Design tasks that test higher-order scientific judgment, including handling ambiguous or messy data, identifying artifacts, selecting appropriate analytical approaches, revising assumptions based on intermediate results, and determining when conclusions are decision-ready; • Build tasks involving realistic scientific data such as FASTA/FASTQ, VCF, BAM/SAM, BED, TSV/CSV, sequence annotations, expression data, and germline or somatic variant data; • Develop tasks requiring multi-step analysis using Python, command-line tools, or established bioinformatics libraries.

We are seeking Chemistry experts to contribute advanced scientific expertise to task design, data curation, and evaluation for frontier AI models and labs. This role focuses on developing rigorous chemistry tasks, reviewing complex scientific outputs, and advising client teams assessing AI-generated solutions.

We are seeking Physical Sciences experts to develop realistic, terminal-based scientific tasks for Terminal Bench Science. You will translate authentic physics, chemistry, materials science, astronomy, and computational science workflows into reproducible benchmark environments.

We are seeking Mathematics experts to develop realistic, terminal-based scientific tasks for Terminal Bench Science. You will design tasks involving numerical analysis, optimization, statistics, mathematical modeling, probability, dynamical systems, and computational mathematics. The role focuses on evaluating whether AI agents can formulate mathematical problems, implement reliable algorithms, operate in a terminal environment, debug numerical workflows, and produce verifiable computational results. Main tasks listed: Design and build authentic computational mathematics tasks — translate mathematical and research workflows into self-contained, multi-step terminal environments, complete with datasets, equations, model definitions, constraints, and expected outputs; Implement expert solutions using Python, R, Julia, C/C++, Bash, or other relevant tools, covering areas such as optimization, numerical integration, differential equations, matrix computation, statistical inference, stochastic modeling, or algorithm analysis; Define rigorous grading criteria, including numerical accuracy, convergence, complexity, feasibility, and mathematical correctness, along with appropriate tolerances, stopping criteria, stability requirements, and reproducibility controls; Develop automated tests and debug issues — validate results across edge cases and alternative valid implementations, and troubleshoot problems involving floating-point precision, solver failures, conditioning, convergence, and performance.

As a Task Executor – Aerodynamics, you will use Flow5 v7 to complete assigned aerodynamic analysis workflows involving foils, wings, aircraft, hydrofoils, and sails. You will receive the required geometry files, project files, operating conditions, and execution steps, and you will follow them exactly. Your work may include modifying geometries, configuring analyses, generating conforming meshes, running Type 1 or Type 2 polars, and exporting the requested results. Main tasks listed: Open and inspect foil, plane, mesh, and project files in Flow5; Build or modify foils, wings, planes, hydrofoils, and other assigned geometries; Configure operating conditions, polar settings, geometry parameters, and mesh requirements; Generate conforming meshes and run Type 1 or Type 2 polar analyses. This work may involve recording or sharing personal data (for example voice, video or account activity): read the consent and privacy terms on the official page before applying.

As a Task Executor – Bioinformatics, you will use UGENE to complete assigned sequence-analysis and molecular-biology workflows. You will receive the required files, values, and step-by-step instructions, and you will execute them exactly as provided. Tasks may include sequence alignment, phylogenetic tree construction, plasmid annotation, workflow execution, and result export. You will save the requested outputs, record your screen when required, and report any ambiguity or mismatch without guessing or improvising. Main tasks listed: Open provided sequence files, plasmid maps, alignments, and tree files in UGENE; Perform instructed operations such as MUSCLE or ClustalW alignments, tree building, plasmid annotation, and workflow execution; Inspect sequence data, annotations, alignments, and analysis results for task-specific verification; Export aligned FASTA files, annotated GenBank files, reports, screenshots, and other requested outputs. This work may involve recording or sharing personal data (for example voice, video or account activity): read the consent and privacy terms on the official page before applying.

We are seeking Life Sciences Subject Matter Experts to develop advanced scientific reasoning tasks for LifeSciBench. Unlike coding-focused benchmarks, this role emphasizes deep biological knowledge, scientific interpretation, and rigorous reasoning. You will create challenging questions based on realistic scientific scenarios, experimental results, biological data, figures, literature-style evidence, and research problems. This is a non-code benchmark: tasks are not executed, and no programming is required. Main tasks listed: Design novel, graduate-level life sciences problems grounded in realistic research contexts; Develop clear questions and rigorous reference answers with defensible scientific reasoning; Interpret experiments, datasets, figures, sequences, structures, and other biological evidence; Validate biological claims, calculations, assumptions, and conclusions.