Science-based AI mineral exploration in Central Asia

Terralign Metals is a mineral exploration company. We identify, rank and file for licences over under-evaluated mineral deposits in Kazakhstan and adjacent Central Asian jurisdictions.

We are not a software vendor. The platform we build is our internal operating system: it exists to answer one question — which piece of ground should we apply for next, and with what confidence — and to carry that answer through to a filed application. In first-application jurisdictions, where a correct filing wins the ground within statutory windows measured in working days, the speed and quality of that answer is the entire business.

Exploration is a data-fusion problem. No single measurement finds a deposit; the signal is the agreement of independent ones.

The archive

Between 1950 and 1991 the Soviet geological programme mapped Central Asia systematically — drillholes, geophysical surveys, classified resource estimates, and the recorded outcomes of complete exploration campaigns. Most of it survives as scanned Russian-language documents no Western explorer has digitised. We read it industrially: OCR, structured extraction, rigorous coordinate conversion, and provenance recorded on every extracted field.

Surface ground-truth

Sentinel-2 and ASTER alteration mapping independently corroborate or contradict the documented record at zero marginal cost per location. Where surface mineralogy and a documented showing agree, conviction compounds; where they cannot agree — buried and covered systems — we say so rather than pretend otherwise.

Subsurface ground-truth

Magnetics and gravity see below cover, where modern discoveries actually sit. Regional grids are fused today; the Soviet programme's high-resolution airborne surveys — recoverable from the archive rather than re-flown — are the planned upgrade from district-scale ranking to target-scale geometry.

The licence clock

National cadastres are monitored live: expiries, revocations and competitor filings surface within the day. Ranked ground is snapped to the statutory licence grid and clipped against allocated and protected land, so a target leaves the system as a fileable block, not a dot on a map.

Everything resolves to one number per unit of ground — a measured, decomposable score with its evidence chain attached. Every claim the system makes traces back to a document page, a satellite acquisition, or a named calculation.

We validate the way we would want to be audited: blind, spatially honest, and with the failures published next to the results.

The negative results are part of the record: experiments that failed our own gates are documented with the same prominence as the ones that passed, because a ranking you cannot falsify is a ranking you cannot trust. Scores are today a validated ordering of ground, not yet a field-calibrated probability — that calibration is precisely what the archive's recorded drilling outcomes exist to provide, and we state the boundary rather than blur it.

One system carries a target from evidence to filing.

• Targets a ranked working environment over six jurisdictions, every score decomposed and auditable
• Cadastre live licence registries, snapshot-diffed; lapse and competitor alerts on any ground we track
• Applications jurisdiction-correct licence bundles rendered from templates, reviewed by geology and legal
• Portfolio held-licence compliance and field data; every result feeds back into the model
• Decision engine a Bayesian belief over each target, updated hole-by-hole, pricing the next survey or metre before it is spent

Single-tenant and internal. Target lists, application drafts and portfolio positions do not leave the platform. Sanctions screening and Soviet-category honesty are enforced at the source: historical estimates are never restated under modern reporting codes without a Competent Person's attestation.

Priority commodities: copper, gold, lithium, rare earths, uranium, molybdenum.