DiMAT Materials Modeler (DiMM): An Interactive Framework for Materials Property Prediction and Optimization Using a Hybrid Machine Learning - Genetic Algorithm Approach

We present DiMM (DiMAT Materials Modeler), an interactive framework that advances materials informatics through an effective integration of machine learning (ML) and genetic algorithms (GA). The framework addresses two fundamental challenges in materials engineering: accurate property prediction and constrained inverse design for small and medium-sized enterprises (SMEs). Our approach uniquely combines dimensionality reduction strategies with an interactive optimization interface, enabling users to efficiently explore material compositions and process parameters while satisfying physical and manufacturing constraints. The framework integrates Large Language Model (LLM) assistants to provide context-aware guidance, making advanced materials informatics accessible to non-expert users. Validation across multiple material systems demonstrates significant improvements in both prediction accuracy and optimization efficiency (e.g., >99% confidence scores and Mean Absolute Error (MAE) ≤ 8.5% for solutions with ≤ 25% free parameters) compared to traditional approaches, while maintaining computation times under two minutes for typical use cases.