FORGE

I built FORGE to explore a question that I think matters as AI becomes more involved in technical decision making. When a model suggests why a machine is behaving abnormally, how do we know whether its explanation is actually correct? FORGE is an industrial investigation platform that brings sensor data, competing explanations, diagnostic experiments, and independent verification into one workflow. The application includes a recorded investigation using Meta's Muse Spark Open model. The model generated three possible explanations for abnormal pressure and sensor readings. After two simulated diagnostic experiments, the independent verifier supported a change in process pressure and a pressure sensor error while finding evidence against normal operation. I also built an interactive 3D pump workstation where users can explore machine components, inspect simulated sensor measurements, capture evidence, run controlled experiments, and review the results. Investigations can be exported as reports for later review. What makes FORGE different is that it does not treat an AI generated explanation as a final diagnosis. It provides a way to test competing explanations against evidence and clearly distinguish supported findings from uncertainty. The current version is a working research prototype using synthetic data. It does not control physical machinery or claim to diagnose real equipment. My goal is to make AI assisted technical investigations more transparent, reliable, and reproducible.