By building a full-process intelligent data system, the Mine Dt has significantly improved the data application efficiency of the mining industry. Targeting the pain points of low efficiency and high cost in data annotation, it has established a mine scenario-level tag system, developed standardized annotation templates by integrating seven major modal data, and realized unified annotation of 50-billion-level sensing data, 3 million video data and 20-billion-token text data, supporting the agile training of small models. Based on the Solstone Model, breakthroughs have been made in cross-modal semantic enhancement technology. An automatic generation mechanism for industry question-answer pairs has been constructed to tackle the problem of long-tail data scarcity, enabling efficient synthesis of fine-tuning data for large models and a leap in the understanding of industry terminology. In parallel, an intelligent conversion system for unstructured text has been developed. Combined with a compliance verification module and dynamic knowledge base partitioning technology, it has realized structured storage and intelligent updating of data such as industry standards and safety regulations, pushing the knowledge retrieval accuracy to over 90%. An integrated, transferable and reusable knowledge base has been formed, fully unlocking the value of mine data assets and providing core driving force for the intelligent transformation of the mining industry.