Physics-informed models
We model the physical fingerprints of ore systems: density and magnetic contrasts, structural corridors and alteration signatures.
Live prospectivity models · South Africa & Montenegro
We map the subsurface, nation by nation.
Our physics-informed AI fuses magnetics, gravity, geochemistry, spectral and structural data. It scores every square kilometre and returns ranked, drill-ready targets in any jurisdiction on Earth.
0km²
of ground scored, cell by cell
0
commodities under live models
0.0
peak model skill (AUC), chromite
0
national deployments in production
WHAT WE DO
We turn open geoscience into ranked drill targets.
LeylineAI is a mineral intelligence company. We fuse every available geoscience layer through physics-informed AI, map where critical minerals are most likely to sit, and send field teams to validate the strongest targets. The method deploys in any jurisdiction on Earth.
Validated on unseen ground
We spatially cross-validate every score on ground the model has never seen. The result is a discovery probability you can drill against.
Operator and platform
We secure ground, run field programs and capture the upside as in-ground value. We hold a stake in every target we rank.
THE DATA
One grid. Every signal.
Every national survey ever flown, sampled or mapped becomes one co-registered grid. Eight independent measurements of the same rock give mineral systems nowhere to hide.
- MAGNETICS
- GRAVITY
- RADIOMETRICS
- GEOCHEMISTRY
- GEOLOGY
- STRUCTURE
- SPECTRAL
- TERRAIN
HOW IT WORKS
From open geoscience to ranked drill targets.
Fuse
We co-register magnetics, gravity, radiometrics, geochemistry, geology, structure, spectral and terrain onto one national grid. Every cell carries every survey that ever touched it.
Model
Physics-informed ensembles of XGBoost, LightGBM and Random Forest models score every cell for the probability of hosting each commodity, constrained by how ore systems form.
Validate
We spatially cross-validate every score on 110 km geographic holdouts, testing on ground the model has never seen. A model that cannot find known deposits blind does not ship.
Target
The strongest cells become ranked drill targets. We secure the ground and validate every one in the field.
“Exploration is a search problem.
We made it computable.”
GLOBAL OPERATIONS
Two live deployments. Three in preparation.
The engine runs wherever open geoscience exists. It is live today across the ancient cratons of South Africa and the Dinaride belt of Montenegro. Italy, Serbia and Bosnia are next.
LIVE OPERATION
PENDING DEPLOYMENT
South Africa
A full national model across the richest mineral endowment on Earth: ancient cratons, layered intrusions and the world's deepest goldfields.
- BUSHVELD COMPLEX
- WITWATERSRAND BASIN
- KALAHARI MANGANESE FIELD
- NAMAQUALAND BELT
Montenegro
The first national-scale prospectivity model of Montenegro's Dinaride thrust belts and karst bauxite provinces.
- DINARIDE BELT
- HIGH KARST BAUXITE PROVINCE
- DRINA-IVANJICA ZONE
- PROKLETIJE OPHIOLITES
Italy · Serbia · Bosnia
Three national deployments in preparation across the Adriatic and Dinaride corridor. Your jurisdiction can follow.
- ITALY
- SERBIA
- BOSNIA & HERZEGOVINA
MINERALS WE TARGET
The commodities that power the energy transition.
We score prospectivity commodity by commodity. Each tile shows the model's spatial cross-validation skill (AUC), a direct measure of how well we separate mineralised ground from background.
AUC 0.5 = COIN FLIP · AUC 1.0 = PERFECT SEPARATION · ALL SCORES ON UNSEEN GROUND
FAQ
Questions we hear from governments, explorers and investors.
What data do you need to deploy in a new jurisdiction?
Any combination of national geoscience: airborne magnetics and radiometrics, gravity, geochemical sampling, geological and structural mapping, satellite spectral data and terrain. The engine is built for the imperfect, mixed-vintage data found in every national archive. More layers sharpen the map. We can start with what exists today.
How is this different from traditional prospectivity mapping?
Three ways. Our models are constrained by the physics of ore formation. Every score is validated on large geographic holdouts, so the skill numbers we publish are measured on ground the model never saw. And we act on our own maps: we secure ground and run the field programs ourselves.
What does the AUC score actually mean?
AUC measures how well the model separates mineralised ground from background on unseen terrain. 0.5 is a coin flip; 1.0 is perfect separation. Our chromite model scores 0.87. Hand it one random mineralised cell and one random barren cell, and it ranks the mineralised cell higher 87% of the time on ground it was never trained on.
Do you license the platform or operate yourselves?
Both. For governments we deliver national prospectivity programs that de-risk licensing rounds and attract investment. For explorers we generate and rank targets over their ground. Where the geology justifies it, we act as an operator, securing licences and running field validation ourselves.
Which jurisdictions are next?
Italy, Serbia and Bosnia are in preparation now. Beyond them, we prioritise countries with strong data archives, under-explored geology and governments motivated to map their critical mineral endowment. If that sounds like yours, talk to us.
GET IN TOUCH
Bring the platform to your jurisdiction.
Government, explorer or investor: if you want a defensible map of what lies in the ground, talk to us. We reply within 24 hours.
contact@leylineai.com