Live · April 2026

Agentic AI for mineral exploration

Joint probabilistic geophysical inversion + 13-agent AI workflow. Two-person team. Built in six weeks. Validated on real public data — Mountain Pass (REE), Bushveld (PGE), and Harrat Rahat (Saudi Arabia).

What we built

Geophysical joint inversion — turning gravity, magnetic, and IP surveys into drill-ready subsurface models — is the analytical spine of mineral exploration. Today it requires a PhD geophysicist and 2–4 weeks per survey. Software like Mira Geoscience and Geotexera either wrap academic codes behind expert-only GUIs or sell inversion as consulting hours.

Tolun AI productizes joint inversion with an agentic layer on top. A geologist uploads survey data, an orchestrator coordinates 13 specialized agents (inversion interpreter, QC agent, drill advisor, report writer, etc.), and the platform returns drill-ready targets in minutes — with full uncertainty quantification, not point estimates.

Real-data proof, not synthetics

Most geophysics demos use synthetic data because real surveys break academic methods. We ran our bounded joint inversion on public gravity gradiometry from Mountain Pass (537,000 stations) and Bouguer gravity from Bushveld (3,877 stations). Cross-sections recover the known ore geometry.

Mountain Pass REE — Bounded 2D Inversion

1,470-station E-W profile · 40×20 mesh · 25m surface cells · alpha cooled 2⁶ → 1 · ±1000 kg/m³ bounds

Mountain Pass — Joint Gravity + Magnetics

1,470 co-located stations · cross-gradient coupled · bounded density + susceptibility · 40–49% uncertainty reduction vs. gravity-only

Bushveld Complex — Bounded 2D Inversion

410-station W-E profile · 680 km extent · 35×18 mesh · PGE reef target · same bounded algorithm

Traction

Commits shipped (last 30 days)

540K+

Real data stations processed

81%

Test coverage (API)

5 / 13

Agents live / mapped

40–49%

Uncertainty reduction (joint vs. single)

Weeks from launch to live product

Why this is hard to replicate

Bounded joint inversion productized

Academic joint inversion papers produce unphysical amplitudes (±4000 kg/m³ on real Bushveld data). We solve it with physically-bounded least-squares on a stacked augmented system. Production-grade, reproducible.

Chat-first geologist interface

Competitors require a PhD geophysicist. We built a chat interface that a field geologist can operate. The agent layer handles the inversion math; the geologist gets drill targets in plain English.

"Epiminds for mining"

13 coordinated AI agents across the full exploration workflow — inversion, QC, drill targeting, reporting. No competitor in mining has this. Epiminds raised $55M Series A applying this exact model to marketing.

Team

Fatimah Abdulghafur, PhD

Co-founder · Geophysics

former Principal Geophysicist at Maaden, $200M programme budget, multi-degree (geophysics, geology, geochemistry), geosciences integrator, international experience across Australia, Canada, USA, Germany, Italy, Kazakhstan, Indonesia, China. Currently engaged with Saudi Vision 2030 mining program. PhD in probabilistic inversion methods.

Brings domain depth + Middle East mining network (Maaden, SGS, MIM, Future Minerals Forum).

Alim Polat, PhD

Co-founder · Engineering · Stockholm

Principal LLM Engineer with 10+ years shipping production agentic AI systems. Just solo-architected SmartDraft at ICON plc — a multi-agent platform with intelligent model routing, RAG over 15,048 documents, and a 4-tier validation framework — in 53 days.

Former AI Lead at Svenska Handelsbanken (7+ years): built and mentored a team of 35, drove fraud detection F1 from 0.3 to 0.926, deployed conversational AI serving 1M+ users, advised C-level on AI strategy. PhD Electronics (Linköping); M.Eng Nanoelectronics (Chalmers). Published in Nature and Cell.

Strategic context

For acquirers evaluating mining-software assets, three natural fits:

Bentley / Seequent

Aarhus + AI

Seequent acquired Aarhus GeoSoftware in 2019 to own the geophysical inversion category. Tolun AI is the 2026 agentic-AI chapter of that thesis — natural tuck-in that adds the AI layer the market is demanding.

Hexagon MineEnterprise

Missing inversion layer

MineEnterprise lacks geophysical inversion. Tolun AI fills that gap and positions Hexagon ahead of Bentley on the AI narrative.

Weir (post-Micromine)

Upstream targeting layer

Micromine handles resource modeling; targeting is upstream of that. Tolun AI makes the Micromine acquisition 10× more strategic by completing the exploration stack.

30-minute technical demo

We'll stream a live inversion on a dataset of your choosing. Mountain Pass, Bushveld, or your own survey. You see the cross-section emerge in real time and the chat agent interpret it for a geologist.