AI Trainer
I generate structured AI training text designed to test how well large language models understand instructions, reason through complex problems, and produce accurate, reliable responses. Much of the content I create consists of carefully engineered prompts that simulate real world tasks, along with detailed scoring rubrics that define how model outputs should be evaluated. These prompts often require multi step reasoning, interpretation of specialised information, or adherence to strict formatting and instruction constraints. I also write high quality reference responses and comparison examples that allow evaluators to assess differences in reasoning quality, factual accuracy, and instruction compliance. A large portion of the text I produce is designed for evaluation and benchmarking. This includes writing prompts that expose common model failure points such as hallucinations, weak reasoning chains, instruction violations, or inconsistencies in logic. I develop clear evaluation criteria that allow responses to be scored objectively, often using atomic and mutually exclusive rules to ensure consistency across assessments. The text I generate frequently incorporates specialised subject matter such as medical knowledge, clinical research processes, nutrition science, and professional decision making scenarios. The goal is to create training and evaluation data that strengthens a model’s ability to perform reliably in complex, real world contexts.