Greater Perth Area
Mining doesn't have a resource problem. It has a decision problem. I'm building GeoVision AI Mining to fix it. After 15 years in economic geology — including AUD 6M+ in research grants, 100+ peer-reviewed publications, adjunct professorships at Curtin University and the University of Queensland, and Stanford's Top 2% Scientist recognition — I left academia to build the AI decision infrastructure the mining industry actually needs. GeoVision AI Mining operates an AI-native exploration system across Australia, China, Zimbabwe, Peru and the Middle East. Our platform, MiningClawd, sits on a 350TB+ multi-source geological data lake and runs four layers of reasoning — Knowledge Nebula → Belt Reasoning Rules→ Post-Training → Agent Applications — turning data into decisions and decisions into mining assets. The deepest constraint in mining is not geological. It's not financial. It's not even technical. It's decisional. AI doesn't replace geologists, engineers, or operators. It amplifies them — letting them see further, decide faster, and validate more rigorously. I write about AI, mining decision systems, and the future of critical minerals. If you're working on these questions — as a mining company, investor, geologist, or researcher — I'd welcome the conversation. 📍 Perth · Hangzhou · Harare 🔗 geovisionaimining.com ✉️[email protected]
Building the mining decision infrastructure for the AI era. GeoVision AI Mining is an AI-native mineral exploration company operating across three hubs: Perth (HQ), Hangzhou & Shanghai (R&D), and Harare (Africa). The company is the institutional vehicle consolidating a decade of AI-in-exploration work I have led with co-founders since 2015 — across three generations of technology: classical big-data systems, CNN-based geological models, and now Transformer-based pre-training / post-training architectures. Our flagship platform, MiningClawd, is a four-layer reasoning system sitting on a 350TB+ multi-source geological data lake: ▸ Knowledge Nebula — a 540-dimension knowledge vector space for mineral systems ▸ Belt Reasoning Rules — distilled from ~2,000–3,000 case-study papers on global mineral systems ▸ Post-Training — vertical commodity models (GV-PorCu, GV-HarLi, GV-LithoSight, GV-GeoStruct) fine-tuned on PhD-level literature ▸ Agent Applications — project-grade agents (incl. our MPM Agent) generating target rankings, drillhole recommendations, and auditable decision evidence chains What sets us apart: ■ Tech-for-Equity, not software-as-a-service. We embed AI into real mining assets and take equity in the outcome — currently active across gold, lithium, copper-gold, and critical minerals projects in Western Australia, Zimbabwe, NSW, Peru, and Saudi Arabia. ■ Active operational portfolio: Malcolm Gold Project, Midlands Gold, Toronto Gold (15% equity), Meta Minerals (NSW), CTZ Gold (WA), Saudi Arabia exploration cooperation (85/15 with WildeSky Resources). ■ A team that combines AI infrastructure depth with ground-truth geology — including Chief Geologist Abbas Ghasemi and CPTO Chen Zekai — plus a research arm (GeoVision AI Mining Research) producing institutional-grade reports on lithium, copper, gold, and the broader capital cycle. ■ Proprietary models trained on a decade of accumulated multi-source data and validated against real drill results.
As Executive Director at AUKT, I spearheaded a strategic pivot to integrate AI-driven exploration methodologies, transforming the company's traditional approach. I led a cross-functional team of geologists and data scientists to deploy our proprietary AI models and Ambient Noise Tomography (ANT) systems. Key results delivered: • Increased Exploration Efficiency: Slashed target identification time by 70%, accelerating the project cycle from years to months. • Enhanced Discovery Success: Improved drilling hit rates by 3x compared to conventional methods, significantly de-risking exploration investments. • Cost Optimization: Reduced overall exploration expenditure by over 40% by minimizing unnecessary drilling and focusing resources on high-probability targets. This successful transformation of AUKT's core operations proved that a data-first, AI-powered approach is not just a concept but a practical solution to achieve unprecedented efficiency and success in mineral exploration. This experience laid the foundational blueprint for the vision we are now executing at LynAI Mines.
Craft and implement strategic plans, assess investment opportunities, establish and coach a local operations team, and oversee the team's execution of exploration projects.
Strategically craft and execute comprehensive business plans, rigorously evaluate potential investment ventures, devise and cultivate detailed strategic business models, assemble and mentor a proficient local operations team, and meticulously supervise the team's implementation of exploration projects to ensure alignment with overarching objectives.
Led research on mineral systems and geochemistry with focus on the Three-River region of Southwest China — one of the world's most significant metallogenic belts for copper, lead-zinc, tin, tungsten, and rare earth deposits. During this period: ■ Supervised PhD and postdoctoral research at the intersection of regional metallogeny and AI-driven prospectivity modelling. ■ Built collaborations with Chinese geological survey institutions and provincial bureaus that later informed GeoVision AI Mining's Greater China research operations. ■ Contributed to the data foundation that now feeds the company's vertical commodity models. Concluded in 2024 to focus on the operational build-out of GeoVision AI Mining and its Series A.
Started building AI-driven mineral exploration in 2015 — three years before KoBold Metals, two years before Earth AI — together with a small group of co-founders who collectively self-funded the project at approximately AUD 6M over a decade. This is the longest continuous AI-in-exploration program I am aware of from a single founding team. It spans three full generations of machine-learning architecture, each rebuilt from the ground up as the technology frontier moved: ▸ Generation 1 (2015–2017) — Classical big-data analytics & global mineral-system modelling. — Multi-source data integration (geophysics, geochemistry, remote sensing, drillhole databases), rule-based prospectivity scoring, and the first version of what would later become our 350TB+ geological data lake. ▸ Generation 2 (2018–2021) — CNN-based geological pattern recognition. — Convolutional architectures for lithology classification, structural feature extraction from multi-resolution geophysical imagery, and the first end-to-end neural prospectivity models trained on Australian and Chinese exploration datasets. ▸ Generation 3 (2022–present) — Transformer-based pre-training / post-training. — The current MiningClawd architecture: a 540-dimension Knowledge Nebula, distilled mineral-system reasoning rules, vertical commodity foundation models post-trained on PhD-level literature, and project-grade decision agents. The technical bet behind this multi-generation rebuild was simple: mining data is heterogeneous, sparse, expensive, and irreversible — and the right AI architecture for it would not be obvious until the broader ML field had matured. We chose to invest through three technology cycles rather than place a single bet, and the resulting data, models, and judgement now sit at the core of GeoVision AI Mining. What this period taught me: ■ Mining AI is not a software problem. It is a knowledge-engineering problem disguised as a software problem.