SLiMS
A Scientifically Rigorous Geological Reasoning Engine
From signal to cited geology.
A reasoning engine for geology — cited, calibrated, and honest about what it doesn't know. Its built core runs calibrated spectral science over a hand-curated knowledge graph and fuses evidence under explicit uncertainty; the natural-language agent that orchestrates these tools and retrieves peer-reviewed evidence is in active development.
A reasoning engine, not a chatbot that has read some geology.
How it reasons
It orchestrates tools — it doesn't guess
A recursive tool-orchestration loop (up to 15 calls), not a single pass.
A hand-curated graph treated as validated truth — 11,419 nodes, 15,854 relationships across chemistry → minerals → rocks → deposit models → spectroscopy.
1,300+ peer-reviewed papers, retrieved as cited evidence (RAG grounding, building) — never the foundation for a claim.
Deterministic, physics-based spectral computation — not curve-fit to one site.
Fuses every line of evidence with an explicit "I don't know" channel. Agreement sharpens. Conflict widens. Nothing is certain alone.
Flags untraceable claims; never fabricates values or citations. When a claim can't be traced, the system declines.
Every answer carries quantified uncertainty, full provenance, and citations. Verified against 36/36 competency questions.
The knowledge graph
Curated truth you can explore
11,419 nodes · 15,854 relationships. Scroll to zoom, drag to pan, click a node to trace its edges.
Decision under uncertainty
Agreement sharpens. Conflict widens. Nothing is certain alone.
SLiMS treats uncertainty as a first-class quantity. Independent evidence that agrees tightens the posterior; evidence that conflicts widens it; and an explicit ignorance channel means no single stream ever reaches false certainty.
Honest build status
Built, building, planned — never blurred
Our credibility with technical and investor audiences is the product. We label every capability.
The reasoning core — curated knowledge graph, calibrated spectral engine, Dempster–Shafer + Yager fusion, the cited-answer contract, and a read-only query agent. Verified 36/36 competency questions.
The natural-language LLM agent / orchestration layer and the publication-corpus RAG grounding.
Full GIS/geophysics ingestion, and in-agent invocation of OreVision and FLUXX as tools.
Map with OreVision. Decide with SLiMS.
The GTA pipeline — pixels → mineral maps → cited decisions.
Who it's for
Built for skeptical, technical audiences
- Government science funders — USGS, DOE national labs, DARPA, NASA — who reward explainability and uncertainty quantification.
- Strategic investors & accelerators — benchmarking defensible, explainable targeting.
- Tech-forward explorers & majors wanting AI-driven targeting they can defend.
- Decision-under-uncertainty research groups.
Cited, calibrated, honest
Detect. Identify. Be sure.
Read the architecture, or talk to our team about partnership and evaluation.