• Hybrid multimodal pipeline using DeepSeek R1, Qwen-Plus, and ChatGPT. • 508 validated mineral records extracted and harmonized into PMMRD. • Supports advanced spatial analysis and evidence-based exploration. • Modular, scalable system for future geological and geophysical integration. Pakistan lacks a centralized, standardized repository for mineral resources, a task traditionally requiring extensive institutional capacity and technical expertise. This study introduces a hybrid multimodal pipeline (MultiMin-MLLM), developed through modern multimodal large language models (MLLMs). The MultiMin-MLLM pipeline integrates heterogeneous geological sources, including scanned geological reports, legacy PDFs, online publications, structured tables, shapefiles, GeoJSON, XML, and OCR-recovered document text, within a unified relational framework. A hybrid architecture integrating DeepSeek R1, Qwen-Plus, and ChatGPT substantially enhances extraction accuracy, semantic coverage, and resilience to noisy or multilingual inputs. The key advantage of the proposed pipeline lies in its ability to support multilingual extraction and harmonization from heterogeneous text-rich geological sources, enabling quick, high-quality extraction from diverse sources, including languages like Chinese, Urdu, Pashto, and others. Manual auditing of key attributes (deposit name, mineral type, coordinates, deposit type) shows field-level accuracy exceeding 92%. The resulting Pakistan Multimodal Mineral Resources Database (PMMRD) contains 508 validated mineral-deposit records, each georeferenced, categorized across geological and administrative dimensions, and fully provenance-linked for traceability. A standardized schema enables spatial analysis, mineral classification, and export to CSV, GeoJSON, and shapefile formats compatible with GIS platforms. Record coverage increased by 27% relative to prior implementations. The proposed framework offers a reproducible foundation for national-scale mineral informatics, supporting multimodal geoscience data integration with geological, geophysical, geochemical, remote-sensing, and other layers. A public Kaggle notebook demonstrates the core hybrid extraction and harmonization workflow. The broader MultiMin-MLLM pipeline is designed to support multilingual and multimodal geoscience data integration across heterogeneous sources, providing a practical basis for expanding mineral data coverage in Pakistan
Bilal et al. (Wed,) studied this question.