Atsushi Hara

Building PromptProof — test LLM extraction prompts against real documents and measure accuracy with statistical confidence. Writing about prompt evals, model upgrades, and LLM reliability.