This paper introduces a novel approach for materials classification using large language models (LLMs) and deep learning. The method leverages prompt engineering to extract textual knowledge from LLMs, which is then used to train a deep learning model for classification tasks. The approach is demonstrated using a dataset of metallic glasses (MGs), achieving a significant improvement in classification accuracy compared to traditional models. The workflow involves defining the classification problem, designing prompts to extract knowledge from LLMs, fine-tuning a BERT model on the generated textual data, and applying the model to classify new materials or study composition-structure-property relationships. The results show that the proposed method achieves up to 463% improvement in accuracy for classifying a specific subset of MGs, highlighting its effectiveness in handling sparse datasets. The method is generalizable and can be applied to various material classification tasks. The study also discusses the potential of integrating advanced language models and domain-specific fine-tuning to revolutionize material classification, especially in scenarios with limited data. The workflow is adaptable to different materials and can be used for tasks such as discovering new materials, analyzing composition-structure-property relationships, and improving model performance through fine-tuning. The paper also addresses the challenges of applying AI to inorganic materials, particularly those with complex structures and large compositional spaces, and highlights the advantages of using textual data generated by LLMs to overcome these challenges. The study demonstrates the effectiveness of the proposed workflow in materials classification and its potential to advance materials discovery and design.This paper introduces a novel approach for materials classification using large language models (LLMs) and deep learning. The method leverages prompt engineering to extract textual knowledge from LLMs, which is then used to train a deep learning model for classification tasks. The approach is demonstrated using a dataset of metallic glasses (MGs), achieving a significant improvement in classification accuracy compared to traditional models. The workflow involves defining the classification problem, designing prompts to extract knowledge from LLMs, fine-tuning a BERT model on the generated textual data, and applying the model to classify new materials or study composition-structure-property relationships. The results show that the proposed method achieves up to 463% improvement in accuracy for classifying a specific subset of MGs, highlighting its effectiveness in handling sparse datasets. The method is generalizable and can be applied to various material classification tasks. The study also discusses the potential of integrating advanced language models and domain-specific fine-tuning to revolutionize material classification, especially in scenarios with limited data. The workflow is adaptable to different materials and can be used for tasks such as discovering new materials, analyzing composition-structure-property relationships, and improving model performance through fine-tuning. The paper also addresses the challenges of applying AI to inorganic materials, particularly those with complex structures and large compositional spaces, and highlights the advantages of using textual data generated by LLMs to overcome these challenges. The study demonstrates the effectiveness of the proposed workflow in materials classification and its potential to advance materials discovery and design.