Advait ΛDVΛIT CoalMindSuper Intelligence · Mining Open live dashboard →
● Operational Intelligence · Opencast Coal

Turn the day's mess into one clear decision.

CoalMind reads the shift reports, WhatsApp, SAP, Coalnet, weather and meetings a mine already produces — and gives the GM an evidence-backed answer to what happened, why we fell short, and what to do next. It never invents a number: what it can't prove, it flags, with a name against it.

⚠ Demo runs on synthetic data — fictional mine, not SECL operations.
14source types unified
0LLM calls in the figures
3timestamps per claim (no future leak)
100%numbers traceable to a source

Six questions, answered every morning

The same loop a mine manager runs by hand — automated, cited, and honest about its limits.

📊

What happened?

Plan vs actual by shift, reconciled across five sources into one headline with a measured band.

🔎

Why the shortfall?

Rain, breakdowns and dispatch delays attributed on the clock — overlaps counted once, never double-counted.

🕑

What did we do last time?

Retrieves similar past incidents, the action taken and whether it actually recurred — evidence, not folklore.

✅

What should we prioritise?

Ranked, approval-gated recommendations that shift with current management directives — safety always first.

⛔

What can't we prove?

Missing or null data is flagged with the responsible role — no plausible-sounding guesses fill the gap.

🔗

How do we know?

Every figure links back to the exact source record, so a manager can audit any claim in one click.

From messy inputs to one answer

One deterministic pipeline, each stage modular, observable and testable.

IngestSAP, Coalnet, shift slips, WhatsApp, meetings, weather, breakdown log
NormaliseFix names (EX3→EX-03), align to date + shift, three timestamps each
ReconcileGroup by lineage → median headline; overlaps resolved on the clock
AttributeLearn tonnes/hour from earlier shifts; keep an explicit residual
RecommendCited, ranked, advisory — a person decides

Full breakdown of sources, trust levels and the timestamp model: DATA_SOURCES.md

Built to be trusted, not just believed

The design rule is simple: admit what you cannot prove.

It shows its uncertainty

On the anchor day, 64% of the shortfall is backed by recorded evidence and 36% is openly flagged as unexplained. A system that explained 100% would be fabricating — the honest remainder is the question worth asking.

It can't see the future

Every output is built "as of" a decision time. A breakdown row that only appears after the repair, or a figure spoken in a later meeting, cannot leak backward into an earlier decision.

It separates fact from estimate

Observed events, statistical attributions and model hypotheses are never mixed. Correlation is never dressed up as cause.

The human decides

CoalMind prepares, ranks and cites. It issues no commands and touches no equipment, blasting or safety-critical action. Approval is always required.

See it work — a day in the mine

A scroll-through of the daily loop on a synthetic mine: eight problems every opencast mine knows, and how CoalMind answers each. Scroll inside the frame; its dashboard button opens the full GM view.

⚠ Synthetic data — a fictional mine, not SECL operations. Open full screen ↗