Software Engineer (Trajectory) (Train AI Models Part Time!)

About this role

*hackajob
is collaborating with
Mercor
to connect them with exceptional professionals for this role.*

• Mercor is hiring experienced Software Engineers\*\* to support high-impact research collaborations with leading AI labs. Freelancers will contribute to building training datasets that improve AI model reasoning and problem-solving on real-world coding tasks. This is a unique opportunity to apply your software engineering expertise toward shaping the next generation of intelligent systems. ### About the Project You'll annotate frontier-model trajectories on \*\*SWE-bench–style tasks\*\* derived from real open-source repositories. Currently, closed-source models do not expose their internal reasoning traces, making it difficult to understand how LLMs approach problem-solving. To address this gap, you'll \*\*reconstruct and annotate the reasoning portions\*\* of model trajectories—using your own problem-solving process and the full task context to infer and infill the underlying thought process at each step. ### Key Responsibilities - Review model-generated code trajectories on realistic software engineering tasks - Reconstruct chain-of-thought reasoning that explains each step of the solution process - Annotate decision points, debugging logic, and problem-solving strategies - Use full task context (codebase, issue descriptions, test cases) to infer plausible reasoning - Ensure annotations reflect realistic developer thought processes and technical accuracy ### Ideal Qualifications - 2+ years of experience in software engineering, with hands-on debugging and problem-solving in real codebases - Degree in Software Engineering, Computer Science, or a related field (Bachelor's minimum; advanced degree preferred) - Strong proficiency in Python, JavaScript, TypeScript, or other common languages found in open-source projects - Familiarity with version control workflows (Git, PRs, issue tracking) - Comfortable articulating technical reasoning in clear, structured w...