Training and evaluating large language model responses through rubric-based ratings, response rewrites, and systematic error analysis for improved model performance.
Large language models need high-quality human feedback to improve. Reviewers must evaluate responses across multiple dimensions — helpfulness, factual accuracy, reasoning quality, clarity, and directness — while ensuring outputs follow required format, tone, structure, and content restrictions.
I review prompts, model outputs, task instructions, and rating rubrics before making judgments. Responses are rated consistently, and when needed, I rewrite or improve outputs to make them more accurate, natural, organized, and professionally useful. Model errors are analyzed with practical feedback to support better training data.
Ratings cover helpfulness, factual accuracy, reasoning quality, clarity, completeness, and alignment with the user request. I verify whether AI answers follow required format, tone, structure, restrictions, examples, and content expectations — flagging partial answers and instruction violations.
Strengthened practical experience in AI quality assurance, prompt review, response evaluation, and remote digital work. Consistent rubric application across writing, coding-adjacent reasoning, data review, and technology support content types.