AI Training Specialist – Agentic Coding
As an AI Training Specialist at Alignerr, I evaluated agentic coding models on multi-step coding tasks and bug fixing using real open-source repositories. I provided detailed RLHF feedback and authored evaluations covering repository navigation, dependency resolution, test execution, and pull-request workflows. I created gold-standard solutions, adversarial prompts, and rubric-based annotations to surface edge cases and model weaknesses. • Evaluated Claude and Cline-based models for code correctness, instruction following, and robustness. • Authored multi-turn agentic evaluations and annotated findings as structured feedback. • Wrote reference solutions for coding tasks to aid in reward model training. • Stress-tested agents under ambiguous specifications, identifying unsafe outputs and reasoning failures.