GlazeSmith : AI Ceramic Glaze Formulation

GlazeSmith is an AI-assisted ceramic glaze formulation tool that lets a potter or ceramicist evaluate a raw material recipe before ever firing a kiln. A deterministic physics engine computes the unity molecular formula and coefficient of thermal expansion using Appen coefficients, producing a real stress-delta against the target clay body - not a guess. Three gradient-boosted XGBoost classifiers, trained on an AMD ROCm GPU instance, predict surface finish, transparency, and color family from that same recipe. A K-NN retrieval layer surfaces the nearest real-world glazes from a reference corpus for comparison, and a Pareto optimizer proposes adjusted recipe candidates. The generated glaze material is mapped directly onto a rotatable 3D vase model so users can visually inspect the fired result rather than reading numbers alone. A conversational agent, powered by Google's Gemma model, lets users ask questions about their recipe in plain language and can verify any suggested change by running it back through the real physics and ML pipeline rather than hallucinating an answer - every suggestion it makes is checked, not asserted.