Projects
This portfolio separates intentionally public research from professional or proprietary R&D. Public projects link to released code, figures, workflows, or research assets. Professional systems are described at capability level only unless ownership and release permissions are explicit.
Public research portfolio
Artificial intelligence-driven metallogenic typing of pyrite
Type: Published research · public-safe repository
Publication: Journal of Geochemical Exploration (2026), 289, 108138
Scale: ~5,200 pyrite analyses · 138 global deposits/settings · six ore-system classes
Methods: RF, SVM, gradient boosting, MLP, class-balance experiments, blind testing, deposit-scale LOGO validation, feature/permutation importance, t-SNE
Public release: notebooks, figures, workflow documentation, reproducibility notes, citation metadata; full compiled global dataset is not publicly released
This project tests whether multielement pyrite geochemistry can discriminate major metallogenic environments more reliably than conventional low-dimensional discrimination diagrams. A key methodological component is deposit-scale validation, designed to reduce optimistic performance from splitting analyses from the same deposit across training and testing.
Big-data galena geochemistry for metallogenic discrimination
Type: Published research · public repository
Publication: Mathematical Geosciences (2026), Early Access
Scale: galena from 37 Pb-Zn deposits and multiple metallogenic classes
Methods: Random Forest, Gradient Boosting, MLP, SVM, standardization, SMOTE, undersampling experiments, blind testing, cross-validation, feature importance, t-SNE
Public release: model notebooks, figures, reproducible workflow; the publisher-protected article is not distributed through the repository
This study evaluates how far galena trace-element chemistry can support data-driven metallogenic classification and which elements contribute most strongly to discrimination across deposit classes.
Gunga Pb-Zn pyrite machine learning
Type: Published research · public repository
Publication: Journal of Geochemical Exploration (2025)
Scientific scope: pyrite trace elements + S-Pb isotopes + global comparison data
Methods: RF, GB, SVM, MLP, deposit-aware/LOGO validation, geochemical and isotopic interpretation
Purpose: test Pb-Zn mineralization and geological-class discrimination while retaining deposit-level geological context
The project connects local Gunga mineral chemistry and isotope data with broader global pyrite patterns to evaluate metallogenic interpretation using both geochemical evidence and machine learning.
Gunga sphalerite deep learning & critical-metal research
Type: Published research · public repository
Publication: Journal of Geochemical Exploration (2025)
Scientific scope: sphalerite geochemistry, S-Pb isotopes, ore-genesis interpretation, critical-metal enrichment
Methods: deep neural networks, global Pb-Zn comparison data, mineral chemistry, isotope constraints, geological interpretation
Purpose: evaluate mineralization style and critical-metal signals, including Ge-bearing sphalerite
This work combines predictive classification with geochemical and isotopic constraints; model output is treated as one line of evidence rather than a substitute for geological interpretation.
Selected professional & computational R&D
GeoAnomalyAI
Status: Active professional R&D · non-public implementation
Role: research-to-product design and scientific validation
Problem: multivariate anomaly intelligence for exploration geochemistry
Capabilities: correlation structure, robust outlier diagnostics, clustering, manifold learning, Isolation Forest-style scoring, spatial confidence, threshold sensitivity, anomaly ranking, and map/target QA
The system is designed to move beyond single-element thresholding toward multivariate, spatially defensible anomaly evidence. Public descriptions intentionally exclude proprietary implementation and client data.
GeoSpectraAI
Status: Active professional R&D · non-public implementation
Problem: multimodal prospectivity screening and evidence integration
Evidence types: remote-sensing/spectral, geological, geochemical, lithological, structural, field, and spatial information
Scientific emphasis: provenance, mineral-system logic, transfer testing, uncertainty, and confidence assessment
GeoSpectraAI is intended to connect remote-sensing and geochemical evidence with geological reasoning rather than treat spectral or ML outputs as stand-alone targets.
Target detection & spatial decision support
Status: Operational / professional R&D · non-public implementation
Problem: translate cleaned exploration data and model outputs into ranked, field-usable targets
Capabilities: geochemical evidence, target ranking, uncertainty products, geostatistical outputs, spatial validation, map QA, batch execution, and reproducible field deliverables
Geostatistical & GPU R&D
Status: Active computational R&D
Problem: scalable spatial uncertainty analysis for mineral exploration
Methods: variography, kriging, sequential Gaussian simulation, uncertainty summaries, sensitivity analysis, Python implementation, and CUDA/GPU experimentation
The focus is not GPU acceleration by itself; acceleration is useful only where deterministic geometry, statistical equivalence, and uncertainty outputs remain validated.
Exploration applications
Saudi Arabia — Arabian Shield
Work: mineral-potential modelling, exploration-screening workflows, target ranking, remote sensing, geological interpretation, and integrated geochemical, geophysical, structural, lithological, and spectral evidence.
Decision context: regional prospectivity, licensing/technical screening, target prioritization, and management-ready exploration outputs.
Mauritania — Sahara Gold
Work: AI-assisted gold targeting using soil geochemistry, remote sensing, geology, anomaly detection, spatial confidence, trench/drill evidence, target ranking, and report automation.
Decision context: evaluate whether multi-source evidence supports specific follow-up targets rather than treating anomaly scores as sufficient evidence on their own.
Tanzania — Mpanda-Mbozi-Karema
Work: multi-element QA/QC, sample-type interpretation, detection-limit and outlier review, spatial-pattern analysis, anomaly confidence, and target-consistency checks.
Decision context: determine which geochemical patterns remain credible after analytical, sampling, and spatial-context checks.
IP and release boundary: proprietary employer, client, or collaboration code, source data, internal thresholds, target coordinates, and unpublished datasets are not made public here unless release is explicitly authorized.
Related pages
Research explains the scientific questions and validation philosophy behind these projects. Publications provides the associated peer-reviewed record, while GitHub contains intentionally released research artifacts.