PrajwalM

**Model Autopsy** is an AI-powered investigation system that helps developers identify data leakage, validation flaws, implementation errors, and other hidden problems in machine learning models. Instead of simply asking an AI to review source files and provide an explanation, Model Autopsy conducts a structured, evidence-driven investigation. It develops competing hypotheses, runs targeted diagnostic tools, evaluates findings independently, and updates its conclusions based on recorded evidence. The system maintains an investigation history, connects findings to supporting diagnostics, and distinguishes verified problems from possibilities that still require examination. It also uses a token-efficient architecture that reuses model context, communicates compact evidence summaries, and enforces limits on iterations and tool calls. Built from my experience developing an AI-powered stock trading model, Model Autopsy addresses a critical question in machine learning: **How can we trust a model's predictions if we haven't investigated whether the model itself is trustworthy?**