Jobs at Gramian Consulting Group

Design and validate reproducible, multi-step computational life sciences workflows for evaluating AI agents. Create datasets, instructions, computational tasks, expert solutions, ground truths, and grading criteria. Execute and troubleshoot scientific analyses, verify code and results, validate assumptions and calculations, and ensure workflows run reliably in isolated environments.

Develop realistic backend coding challenges, including feature implementations and bug fixes, for AI training projects. Design deterministic verifiers, automated tests, and validation criteria that distinguish correct from incorrect solutions. Evaluate code for correctness, reliability, maintainability, security, and performance while identifying edge cases and bottlenecks. Work independently with backend technologies and refine tasks based on project requirements and quality standards.

Debug and improve diverse codebases by fixing software defects, identifying vulnerabilities, conducting security assessments, refactoring legacy code, and optimizing secure backend systems. The specialist will document technical findings and contribute cybersecurity and software engineering expertise to AI training datasets. Work is fully remote and requires independent execution across project specifications.

Designs challenging computational genomics and bioinformatics tasks for AI benchmarking. Develops reproducible Python-based workflows using biological datasets, command-line tools, and bioinformatics libraries; creates reference solutions, ground truths, grading criteria, and documentation. Validates scientific accuracy, diagnoses data artifacts, assesses analytical outputs, incorporates reviewer feedback, and communicates technical requirements to project stakeholders.

Design rigorous physics coding tasks for AI evaluation, including connected subproblems, verified Python solutions, unit tests, and discriminative test cases. Perform quality checks using the Turing Central Task Platform, revise work based on feedback, and optimize tasks for LLM evaluation criteria. Validate scientific correctness, determinism, well-posedness, and expected outputs while participating in reviews and standups during PST overlap hours.

Design difficult open-web research problems for benchmarking frontier AI browsing agents. Responsibilities include constructing multi-step questions, validating clues through government and institutional sources, tracing evidence across archives and PDFs, documenting auditable citations, testing search difficulty, eliminating ambiguity, and delivering reproducible research documentation.

Create rigorous mathematics coding problems, verified Python solutions, unit tests, and discriminative evaluation cases for frontier AI models. Perform quality control through the Central Task Platform, refine tasks based on feedback, and optimize datasets for LLM evaluation. The role requires advanced mathematics expertise, scientific programming, deterministic outputs, mathematical research or writing experience, and familiarity with AI-generated code assessment.

Research Fellows will validate AI-generated paper reproductions, simulations, and results; map research taxonomies; define standards for high-quality AI-assisted science; and create benchmark tasks and evaluations for frontier AI models. Fellows will contribute to a publicly released benchmark paper and open evaluation set as lead or first authors. The contract offers flexible, remote global work for current STEM PhD students or postdocs with published research and hands-on AI agent experience.

Conduct advanced research in Algebraic Geometry, develop and critique complex mathematical proofs, investigate foundational problems, and build theoretical frameworks. Formalize structural patterns into reusable specifications and technical models, contribute to mathematically inspired architecture design, communicate abstract concepts to technical audiences, document findings, and collaborate with a specialized research team. This is a four-week contractor assignment requiring PhD-level expertise and a strong publication record.