HeisenBig

FusionPilot is an AI-powered magnetic-design laboratory that makes frontier fusion research easier to explore and understand. It investigates a concrete question: can tolerance-aware optimization preserve magnetic-field accuracy when coils are built slightly differently? Users explore interactive 3D coils, define construction tolerances and engineering constraints, and ask Meta Muse Spark to propose an experiment. The model returns a hypothesis, bounded settings, and a verification plan. FusionPilot validates those settings before a first-principles physics solver evaluates them. The application calculates vacuum magnetic fields using Biot–Savart integration and compares nominal-only and tolerance-aware searches under the same field-evaluation budget. Independent perturbation samples, a finer validation mesh, and repeated random seeds help users examine performance beyond the search conditions. Results include field error, variability, constraint failures, coil geometry, and search history. Meta interprets the measured evidence, while exportable configurations preserve settings, seeds, results, and model provenance for reproducibility. Built with TypeScript, React, Three.js, and background numerical workers, FusionPilot connects natural-language investigation with inspectable scientific evidence. This is an exploratory vacuum-field prototype, not a full plasma simulation or reactor design. It reports negative results honestly and does not assume that robust optimization always wins.