Language-Augmented Semantic Priors for B-Spline Surface Fitting

Conventional CAD geometric kernels typically rely on heuristic initialization for B-spline surface fitting, discarding procedural design intent embedded in modeling histories. Rather than altering core numerical solvers, the proposed LASP framework acts as an external reasoning layer. It translates procedural modeling logs into structured text, using a fine-tuned large language model to predict executable B-spline prior parameters. By injecting semantic guidance into classical solvers, the system demonstrates how language-driven reasoning can serve as an inductive bias for precision geometric optimization.