EarthScience.AI Neurons AI Pvt Ltd
We build the intelligence layer that runs the full length of the mineral value chain — exploration, mining, mineral processing and metallurgy — and makes the uncertainty in every decision visible, quantified and auditable.
A single smoothed surface fitted through the sample points. It is tidy, confident and wrong in ways nobody can see. Smoothing suppresses real variance and invents anomalies in the gaps between data.
A realization that honours every one of the same measurements exactly — and looks nothing like the smooth map. It keeps changing because the data genuinely does not decide between these outcomes.
Spread across the full ensemble. Dark is constrained by measurement; bright is interpolation dressed as knowledge. This is the layer that should sit under every decision downstream — and almost never does.
A note on what you are looking at. Every visual on this page is a parametric illustration of method, computed live in your browser from the controls you move. There is no client data here, no case-study figures, and nothing fabricated to resemble a result. We do not publish other people's ground.
01 The problem
Every serious decision in this industry — fly the survey, sink the hole, declare the resource, draw the dig line, change the setpoint, sign the offtake — is a bet placed on a picture of the orebody or the process that was assembled from a handful of measurements and a great deal of interpolation.
The measurements are excellent. The interpolation is where the money goes. A geochemical survey samples a few thousand points across a district and a smoothing algorithm fills in the rest. A drill programme intersects a few thousand metres of a deposit that is hundreds of millions of tonnes. A shovel operator follows a dig line drawn from blast hole samples on a grid coarser than the ore itself. A concentrator is instrumented to the second and tuned to the ore it was commissioned on. A smelter contract prices a concentrate whose variability was never characterised.
In each case the industry's standard output is one number — one surface, one block model, one dig line, one setpoint, one grade — presented without the range of other answers that fit the same evidence just as well. The confidence in the deliverable is a property of the rendering, not of the data.
That gap does not stay in one place. It compounds. An over-smoothed anomaly becomes a wasted drill programme. An over-optimistic block model becomes a dig line that sends ore to the dump. Feed that nobody characterised becomes a circuit running at the wrong setpoint. A concentrate whose deleterious elements were never modelled becomes a penalty clause. By the time it reaches the reconciliation report it is a number nobody can explain.
EarthScience.AI exists to close that gap along the entire chain. Not by producing prettier maps, but by attaching a quantified, auditable confidence to every number a mining company acts on — and turning that confidence into a decision.
02 Four domains, one problem
These are four distinct professions with four vocabularies, four sets of software and usually four reporting lines. They share one structural weakness: each acts on a single estimate whose error is known to nobody, and each passes that error downstream disguised as a fact.
Finding it. Regional ground selection, anomaly detection, target generation, drill programme design and resource definition — under cover, at depth, with falling discovery rates and rising search costs.
Getting it out. Pit and stope design, cut-off strategy, grade control and dig lines, drill and blast, fleet and haulage, geotechnical risk, stockpiles and reconciliation.
Concentrating it. Feed presentation and blending, comminution, classification, separation and concentration, dewatering, water balance and tailings — plus cement and pyroprocessing lines.
Turning it into metal you get paid for. Metallurgical domaining, leaching and hydrometallurgy, solvent extraction, smelting and refining, impurity deportment, residues, and the offtake terms that price all of it.
03 Domain one
The shallow, outcropping, easily-found deposits are gone. What remains is deeper, blinder, lower grade and under cover — while the critical minerals build-out has compressed the time available to find and permit new supply. The industry cannot drill its way out of that. Each survey and each hole has to carry more information than it used to.
—
The same arithmetic decides the next hole, not just the next survey — and it is the only rigorous basis for a stopping rule. Sequenced hole by hole against an updating picture of the orebody, it is the difference between a campaign that ends when it has learned enough and one that ends when the money runs out.
04 Domain two
Every tonne moved is a decision that cannot be taken back. The pit design fixes what is economic for a decade; the dig line fixes what is economic for the next four hours. Both are drawn from estimates — and between the two of them sits the largest and least-examined value leak in the industry.
The real grade, block by block. Nobody ever sees this — it is reconstructed months later, badly, from mill reconciliation.
The dig blocks as grade control classifies them, from the blast hole samples shown as dots.
Why this is invisible in the monthly report. Misclassification concentrates at the cut-off, where the margin per tonne is near zero — so in margin terms these errors look almost free, which is exactly why they survive scrutiny. In metal terms they are not free at all: the tonnes on the dump are gone permanently, and the waste in the mill consumed capacity that ore could have used. The reconciliation factor in the third box is usually the only visible symptom.
05 Domain three
A concentrator or a cement line is designed once, against a design ore, and then runs for thirty years against ore that is never the design ore again. Hardness, liberation, clay, moisture, oxidation and feed grade all drift. The economic optimum drifts with them, hour by hour. The control philosophy does not — so the plant runs to stability, and the gap between stable and optimal recurs every shift and requires no capital to close.
Move the disturbance sliders. That is the ore changing underneath the plant — the thing nobody scheduled, and the thing the control room usually learns about from the assay lab, hours later.
—
—
Calcination is stoichiometric — decomposing the carbonate that supplies clinker CaO releases about 525 kg CO₂ per tonne of clinker. Fuel emissions come from the heat you enter. Whole-plant electricity is held at 100 kWh and 0.72 kg CO₂/kWh; adjust both to your own grid before quoting anything.
—
Calcination CO₂ is released by the limestone decomposing, and it is fixed by the chemistry — burn cleaner fuel all you like, it does not move. Drag the clinker factor and watch what happens. Carbon exposure in cement is a raw mix and product design problem long before it is an energy problem, and it is now priced at the border.
06 Domain four
Everything upstream is measured in tonnes and grades. Metallurgy is where those become payable units — and the conversion is far leakier than most operations acknowledge. Some of the loss is chemistry you can improve. Some is a contract clause you can negotiate. Knowing which is which, per shipment, is the whole discipline.
Ore grade is held at 0.60% for this illustration so the cascade reads as percentages of contained metal rather than as one deposit's economics. Payable metal follows the common convention of the lesser of a fixed unit deduction and a percentage payable; penalty thresholds and scales differ by smelter and are negotiated, not universal.
07 Doctrine
A grade surface, a magnetic grid, a block model — each is an estimate, and an estimate without a spread is not a result, it is an opinion with a colour scale. The question is never "what is the value here", it is "what is the probability the value here exceeds the threshold that changes my decision".
Every surface we produce carries its own uncertainty, and every recommendation is expressed as a probability against a threshold the client sets.
What is left is under cover, under water, or under a hundred metres of transported regolith that has erased every surface expression. Surface geochemistry cannot find it. Mapping cannot find it.
It is found by recovering the physical properties of the rock — density, susceptibility, conductivity, chargeability — from geophysical response, in three dimensions, with the non-uniqueness of that recovery stated openly rather than hidden inside a single preferred model.
Grade control draws a line between ore and waste using an estimate built from blast hole cuttings — a biased sample — on a grid coarser than the grade actually changes. Then a shovel with a defined bucket follows that line at night, in dust.
Everything lost to that mismatch is lost permanently: ore on the dump is never recovered, and waste in the mill is paid for twice — once to haul it and once to grind it. This is the largest routine value leak in mining, and it is almost never quantified against what perfect selection would have delivered.
When the mill says something different from the model, the industry's reflex is to explain the variance in the monthly report. The variance is information: it is the orebody telling you which assumption in the estimate was wrong, and roughly by how much.
A reconciliation gap that recurs in the same direction is not noise and not bad luck. It is a correctable bias, and every month it goes uncorrected it is being paid for again.
Hardness, liberation, clay content, oxidation and feed grade drift shift by shift, and the setpoints that maximise value drift with them. But control philosophy is set at commissioning, tuned to a design ore, and defended by habit long after the deposit has changed underneath it.
The recoverable margin between "stable" and "optimal" is usually larger than anything available from new capital equipment — and it is available now, from instrumentation already installed and already logging.
Replaying live tags in a 3D model tells the operator what they can already see out of the window. The value is entirely in what it does beyond the mirror: forecasting the state the circuit is heading into, naming the binding constraint, and putting a number on the cost of staying where you are.
And it has to span the circuit. A grinding model that does not know what the flotation cells will do with a coarser product will happily push throughput straight into the tailings dam — and report success while doing it.
Plant recovery is reported daily and defended fiercely. It is one term in a chain that also contains unit deductions, treatment and refining charges, moisture, freight and penalty elements — and several of those are larger than the recovery improvements being fought over upstream.
The discipline is not maximising recovery. It is maximising payable units net of every charge, which sometimes means running the circuit somewhere a recovery-only KPI would call a bad day.
Arsenic, antimony, bismuth, fluorine and mercury arrive in the concentrate because they were in the ore, and they arrive in pulses because mineralisation is zoned. By the time they show up in the assay of a loaded shipment, the blending decision that could have fixed them was taken weeks earlier in the pit.
Predicting impurity deportment from the resource model, and blending against it before it becomes a penalty, is worth more than any downstream cleaning circuit and costs a fraction as much.
08 Standing
We do not demonstrate on toy datasets. What follows is the scale we have actually worked at — the methods and the deployments themselves stay behind the NDA, where our clients expect them to be.
09 The ground we work on
A model a competent person cannot sign, a regulator cannot accept, or a buyer cannot verify is not an asset. Our deliverables are built against the codes the industry is actually held to.
10 How an engagement runs
We take the decisions you are actually about to make — this survey, this pushback, this circuit, this shipment — and identify which are currently being made on unquantified confidence. The output is a ranked list of where the leverage is, and it sets the scope of everything after it.
One deposit, one bench, one circuit, one concentrate stream — whatever the audit ranked first. We run the full scope through to a defined success criterion agreed before we start, rather than stopping at intermediate milestones and leaving the study half-answered.
Outcomes are delivered as one consolidated package, with the capability moved into your environment, your team trained on it, and every output traceable back to source data. Your ground, your models, your IP — under your governance.
No format is refused. Instrument-native, proprietary, legacy, scanned or hardcopy — conversion, digitisation and georeferencing are done in-house and are part of the programme, not an obstacle to it. You will not be told after the fact that something could not be processed.
That includes the material most operations assume is unusable: decades-old drill logs in filing cabinets, survey data in formats whose software no longer exists, marked-up plans, historian exports nobody has opened, and laboratory records held as paper. In our experience the cheapest metre of new information a company can buy is usually one it already paid for.
EarthScience.AI is a pure artificial-intelligence and machine-learning platform. We do not run field campaigns, deploy on-site crews or supervise drilling, and we do not carry out physical plant engineering. We sit above all of that — over the data your own teams, contractors and instruments generate — and turn it into decisions that can be defended in a board meeting, an audit or a data room.
Where physical execution is required, it is delivered by our partner GeoExpOre. Ground exploration — field mapping, surveying, sampling campaigns, trenching and drilling supervision — and plant design for mineral processing are handled by GeoExpOre.com, a mineral exploration consultancy and Notified Private Exploration Agency. That means a programme can run end to end — intelligence layer and boots on the ground — without you having to assemble it from two unrelated procurements.
The two remain separate organisations with separate scopes, contracts and accountabilities. You can engage either on its own.
11 Start the conversation
One target you are unsure of. One programme you cannot justify stopping. One bench where the reconciliation never closes. One circuit leaving margin in the tailings. One concentrate that keeps attracting penalties. Tell us which, and we will show you how it looks with the uncertainty attached.