ERITY will present at the upcoming Australian Institute of Geoscientists ML/AI Applications in Greenfields Mineral Exploration workshop delivered in partnership with Geoconferences WA.
The event will be held at the University Club of Western Australia on 29 July 2026, bringing together practitioners from industry, academia and government to explore the practical application of machine learning and artificial intelligence across the greenfields exploration lifecycle.
Lead AI Engineer, Laura Jiang, will present GeoAgent: An Agentic Vision-Language Architecture for High-Fidelity Legacy Data Synthesis in Greenfields Exploration.
The workshop is scheduled to be held at the University Club of Western Australia on 29 July 2026.
The presentation will demonstrate how GeoAgent applies agentic AI and vision-language models to transform legacy geological reports, maps and technical documents into structured, searchable knowledge, supporting more informed and defensible exploration decisions.
Participants will gain insights into real-world ML/AI workflows, practical implementation approaches, successful and unsuccessful case studies, and the opportunities and limitations of AI in mineral exploration.
The workshop provides an important forum for sharing knowledge, fostering collaboration, and advancing the responsible application of AI across the geoscience and resource sector.
Abstract
Geoscience organisations manage decades of legacy reports and geological maps containing critical (sub)-surface knowledge. However, this information is largely stored in unstructured and scanned formats, limiting accessibility and significantly constraining mineral exploration, especially in greenfield environments. Standard Retrieval-Augmented Generation (RAG) pipelines often treat these documents as flat text, discarding the structural hierarchy and spatial continuity essential for interpreting geological maps and cross-sections.
At ERITY, we have developed a structure-aware agentic RAG system, called GeoAgent, that integrates the Docling parser for layout-aware ingestion, enabling reliable extraction of structured content such as geochemical tables, section hierarchies, and multi-column reports. It further incorporates Gemini-1.5-Flash for automated visual context extraction from maps. The workflow is orchestrated via a state-based pipeline, featuring self-reflective nodes for retrieval grading and hallucination checking. GeoAgent is evaluated on a dataset of legacy geological reports, comparing its performance against standard fixed-window RAG baselines.
Preliminary analysis focuses on answer faithfulness, correctness in stratigraphic context, and the retrieval of multi-page spatial assets. Our results demonstrate that preserving document structure and leveraging vision-language models for map indexing helps reduce hallucination and improves the retrieval of complex spatial data. This architecture provides a scalable path to digitising institutional knowledge in the geosciences, ensuring that spatial and structural relationships are preserved for downstream analysis and decision-making, particularly in greenfields.
ERITY Speaker
GeoAgent: An Agentic Vision-Language Architecture for High-Fidelity Legacy Data Synthesis in Greenfields Exploration
Session 3: Legacy Data, AI Agents & Geological Reasoning | Date: Thursday, July 29 | Session Time: 14:50 hrs. – 15:15 hrs. | Presentation Time : 14:00 hrs. – 16:00 hrs.
Laura Jiang
Lead AI Engineer
Laura is a Lead AI Engineer at ERITY, specialising in applied artificial intelligence for complex technical and geoscience datasets.
Her work focuses on agentic AI, multimodal document intelligence, retrieval-augmented generation (RAG), and vision-language models, transforming unstructured technical information into actionable knowledge. She recently completed a Master of Computing at Curtin University and contributed to published research on heterogeneous graph neural networks for cybersecurity anomaly detection.
Prior to joining ERITY, she gained experience in computer vision and machine learning through projects with DJI and Honda. Her interests include document intelligence, multimodal AI, and AI-enabled decision support for the mining and resources sector.