Contribute to a pioneering project focused on evaluating frontier models by designing and validating complex benchmark tasks in data science, machine learning, finance, and coding. This role emphasizes the development of robust, real-world tasks with executable tests, followed by the analysis of model and agent behavior to identify reasoning and problem-solving gaps.

• Design challenging, real-world STEM problems.
• Implement each task within an agentic development environment using Python.

• Deep expertise in data science, machine learning, finance, and/or Python-based coding.
• Active or recently graduated PhD from a top U.S.-based school.
• Strong research background in frontier STEM topics.
• Ability to engage reliably for 30+ hours per week, primarily on weekdays.
• Demonstrated technical output, such as high-quality open-source contributions, especially in agentic or LLM tooling ecosystems.
• Comfort with reading and reasoning about agent behavior traces to diagnose failure modes beyond surface-level errors.

• Initial focus area: agentic workflows for STEM tasks.
• Familiarity with agentic frameworks and OSS ecosystems is beneficial (e.g., LangChain, MetaGPT, AutoGen, AutoGPT, CrewAI, LlamaIndex, BabyAGI, SuperAGI, CAMEL, AgentGPT, Dify).
• Deliverables are expected to be reproducible and testable, with clear specifications, deterministic tests where possible, and documented environments.

is a talent marketplace connecting top experts with leading AI labs and research organizations. Our investors include Benchmark, General Catalyst, Adam D’Angelo, Larry Summers, and Jack Dorsey. Thousands of professionals across various domains, including law, creative fields, engineering, and research, have joined to work on groundbreaking projects shaping the future of AI.