Deterministic engine, not LLM math
Resource estimates, mine plans, cost builds, NPV/IRR/DCF and sensitivities are computed by a versioned, deterministic engine. The same inputs always produce the same outputs — reproducible, diffable, auditable.
Our technology
AI writes the words. It never writes the numbers.
Every figure comes from a deterministic engine; the language model is grounded by retrieval and checked by humans. That split is the whole point.
RAG + LLM report writing
A retrieval-augmented pipeline grounds the language model in your project data and a curated standards corpus. The AI writes prose and structure; it is never the source of a number.
Three-layer no-error guarantee
Layer 1: the deterministic engine owns every figure. Layer 2: golden-dataset regression tests catch any drift on every release. Layer 3: a qualified human reviews and signs off before export.
Standards-compliant by construction
Documents are generated against the structure regulators and lenders expect — built around Mongolian TEFS practice and resource-classification standards, with international reporting conventions in mind.
Architecture
How a study flows through Baku AI
One directed path from source data to signed document — no step where a number can appear out of thin air.
Drill logsAssay tablesSurveysLab certificates
Ingestion & QA
Schema validation · unit checks · outlier flags · anonymisation
Deterministic Core
Resource model · mine plan · costs · NPV / IRR / DCF · sensitivity
Pinned by golden-dataset regression tests on every release
RAG Knowledge Base
Standards corpus · prior studies · domain terminology
LLM Report Writer
Writes narrative & structure — numbers are injected from the core, never generated
Human Sign-off
Qualified expert reviews and approves
Export
Word · PDF · Excel appendices
Every figure in the document traces back to a source row and an engine version.
Engine output
Built to be interrogated
Sensitivity, scenarios and market context ship with every study — the kind of output a lender's analyst pulls apart first.
NPV sensitivity — tornado
ΔNPV ($M) when each driver moves ±15%
Illustrative sample data
Commodity price outlook
RoadmapPrice index, 2022 = 100 · dashed = model forecast with p10–p90 band
Illustrative sample data
Roadmap
Commodity intelligence: prices, not just projects
The next layer of Baku AI predicts the market your project sells into — AI-driven price forecasting for copper, coal and other mining commodities, with scenario feeds straight into your financial model.