<p>Thermotropic liquid crystals (LCs) underpin technologies from adaptive optics to responsive soft materials, yet their design is constrained by the difficulty of predicting two key phase transition temperatures—the melting temperature and clearing temperature—across structurally diverse subclasses. Here, we present the first unified, open-access LC prediction platform integrating large-scale data curation, LC/non-LC classification, and phase transition regression. Using a curated dataset spanning rod-like, discotic, and bent-core LCs, we trained an LC-favoring majority-vote ensemble model that maximizes recall, virtually eliminating false negatives in candidate screening. For melting temperature prediction, an ensemble of random forest and a graph neural network based on the message passing neural network architecture achieved the highest accuracy, while for clearing temperature, the graph neural network alone provided sufficient generalization. The precision of the framework reproduces even subtle odd–even effects and identifies structural motifs underlying high-error outliers, offering mechanistic insights into mesophase stability. Deployed as the Liquid Crystal Predictor web tool, this platform enables data-driven LC property prediction for community usage and establishes a scalable route toward data-driven discovery of advanced LC materials.</p>