NUMBIRD
Above expectations. Beyond possibilities.

We see what the world
is blind to.

Autonomous drones, 26 years of satellite history and 29 machine-learning models in production — fused into one predictive intelligence engine. 127.7 million records already processed across ten countries.

MineralScanPetroScan AgroScanFloodScan DisasterScanLogisticScan MXchange

NUMBIRD · MARKUS IT LLC · Florida, USA
Confidential — prepared for our partners of BRISA America

The problem, stated plainly

The data already exists.
Nobody can act on it.

Every mineral agency, land registry, environmental monitor and satellite constellation on earth publishes data. It sits in incompatible formats, in different languages, at different scales, updated on different clocks. The industry's answer has been to hire people to read it.

What that costs you today
  • Geologists spend weeks assembling a picture that changes monthly
  • Title, royalty and environmental data live in separate systems that never reconcile
  • Field campaigns are commissioned on intuition, then justified afterwards
  • By the time a report is written, the satellite pass is stale
What we built instead
  • Continuous ingestion — the record updates itself, daily
  • One schema across nine countries, three languages, 51 datasets
  • Models that rank ground rather than describe it
  • Drone tasking closes the loop: the machine chooses where to fly next
The one-line version

We are not a data vendor and not a consultancy. We built the instrument — and it is already running in production on 127.7 million records, not in a pitch deck.

Ninety seconds

See the fleet at work.

Prefer YouTube? youtube.com/watch?v=_X1D1B9DSHQ

thenumbird.com · the film is public, the platform behind it is not

The architecture

Three layers. One engine.

Each layer is useless alone. Competitors sell one of them.

01
Earth observation Sentinel-1 and Sentinel-2, Landsat, MODIS, Copernicus and SRTM through Google Earth Engine. Twenty-six years of history, so change is measurable rather than asserted — and the whole territory is under watch before anything is tasked.
02
Autonomous aerial fleet LiDAR, hyperspectral and thermal at 100 m — resolution a satellite physically cannot reach. The fleet flies where the orbital layer raised a flag, on a mission plan the model wrote, not a pilot.
03
Prediction & the data engine 29 models in production — graph neural networks, XGBoost, boosted trees, ARIMA and anomaly detection — over 127.7 million records on BigQuery and Vertex AI. They rank ground, score risk, price assets, and decide what the fleet flies next.
The part nobody else has closed

The loop runs without a human in it. The satellite layer flags an anomaly, the model scores it, the drone is tasked to fly it, the imagery returns and re-trains the model. Plenty of companies fly drones. Plenty sell satellite analytics. We know of none that let the machine decide where to fly next.

Sources · Copernicus Sentinel-1/2, USGS Landsat, NASA MODIS and SRTM via Google Earth Engine · NumBird BQML / Vertex AI model registry, 26 Aug 2026.
Not a roadmap — a warehouse

What is already running.

Measured against the live platform on 24 August 2026. Every figure below is a query result, not an estimate.

127.7M
records processed
1,247
tables in production
29
ML models deployed
9
countries with live data
PlatformDomainRecordsStatus
LogisticScanInfrastructure & transport41,581,417Production
FloodScanFlood & wetland risk · USA38,293,686Production
MineralScan BrazilMineral intelligence18,505,800Production
DisasterScanDisaster & deforestation15,452,905Production
AgroScanAgriculture6,144,540Production
MineralScan LATAM + Africa8 further countries5,435,858Production
PetroScanOil & gas1,234,469Production
MineralExchangeAsset pricing & market1,038,460Production
Read that table again

The largest system is not the mineral one. The same engine ingests roads, floods, crops, fires and wells — because the hard part was never the mineral domain. It was the pipeline. Point it at a new registry and a new country goes live in weeks.

Source · NumBird production warehouse on Google BigQuery — __TABLES__ row counts and the BQML model registry, queried 26 Aug 2026. Every figure is a query result, not an estimate.
Who executes

The squad.

Three executives resident in the United States, each individually examined and approved by USCIS. Three separate processes, three approvals.

Marcos Simplicio · CEO & Architect

EB-2 National Interest Waiver — exceptional ability in AI. Built 10 proprietary platforms from zero with no external capital. 15 years in data, AI and cloud across three continents; led squads at Big Four and S&P 500. GE · Deloitte · Petrobras · Bunge · Renault · Carrefour.

Alexandre Prado · CFO

EB-1 permanent resident. Finance, corporate structuring and capital discipline — the function that keeps a platform business honest about its unit economics.

Alexandre Zocche · COO

US citizen. Twelve years in the United States, recruited into the country by Eli Lilly. Operations, delivery and the commercial relationships that open doors.

Why this is on a slide

Ten platforms in production, built by three people without outside money, is either a claim or a demonstration. Everything in this deck can be opened in a browser in front of you — that is the difference.

Sources · USCIS immigration classifications (EB-1, EB-2 NIW) · employer records, Eli Lilly · public company filings.
Why Brazil, and why now

The ground the world needs
is mostly here — and barely mapped.

Brazil holds the second-largest rare-earth reserves on earth and supplies roughly nine tenths of the world's niobium. What follows is not market commentary: every figure is measured in our own copy of the federal register.

R$ 54.4B
iron royalties — 73% of all CFEM
2,574
rare-earth claims · 22 states
229
entities hold every one of them
4.13M ha
rare-earth ground under title
The critical-minerals position, measured
  • Rare earths — 2,574 claims across 22 states, 4.13M ha, 229 holders. Concentrated in Bahia (1,006), Minas Gerais (926) and Goiás (496).
  • Lithium — 4,163 claims, 5.88M ha, 559 holders. The "Lithium Valley" of Minas is inside this register.
  • Niobium — 550 claims, 1.23M ha, 153 holders, in the country that supplies ~90% of world output.
  • Gold — R$3.7 bn in royalties across 87,831 separate payments: a long tail no incumbent tracks.
What that concentration means
  • The strategic question of the decade — who controls non-Chinese rare earths — resolves in Brazil to a list of 229 names. We hold that list.
  • Iron is 73% of royalties but a fraction of the titles: the register's value and its volume sit in different places.
  • Gold's 87,831 payments against iron's 33,710 is the signature of a fragmented, under-consolidated field.
  • None of this is inferable from a market report. It comes from the title register, the royalty ledger and the holder behind each one.
Sources · ANM SIGMINE cadastral register and CFEM royalty series 2002–2026 (dadosabertos.anm.gov.br), synchronised 26 Aug 2026 — 2,382,446 royalty records, 100% parsed · reserve and world-share context: US Geological Survey, Mineral Commodity Summaries 2026.
The flagship · Brazil

Every mining claim in Brazil,
scored.

We ingested the complete register of the Agência Nacional de Mineração and rebuilt it as an intelligence layer — title, substance, phase, polygon, ownership and royalty history.

254,715
concessions scored
R$ 74.5B
royalties tracked · 2002–2026
3,820
municipalities
316
substances
What sits underneath
  • 2,382,446 individual royalty payments, reconciled to claim and municipality
  • 229,237 claims resolved to their legal holder — 55,476 distinct owners
  • 444,908 Amazon deforestation alerts cross-referenced against concession polygons
  • 309,499 regulatory process records — every filing, every phase change
  • 27 states, one schema, updated continuously
What the models do with it
  • Rank exploratory potential per polygon, not per region
  • Flag regulatory risk before it becomes a stoppage
  • Price the asset from its own production and royalty record
  • Detect anomalies between what is declared and what the satellite sees
Sources · Agência Nacional de Mineração (ANM) — SIGMINE cadastral register and CFEM royalty series (dadosabertos.anm.gov.br) · INPE/PRODES deforestation alerts · scored on NumBird models.
The ownership graph

A title is a number.
We give you the owner.

The public register lists a process number and a polygon. It does not tell you who stands behind it, what else they hold, or where. We resolved the legal holder for 229,237 concessions and rebuilt the register as an ownership graph — so a counterparty, a competitor or an acquisition target can be seen whole.

229,237
claims resolved to a holder
55,476
distinct owners identified
3.6%
held by the top ten
27
states, one ownership schema
The largest holders, measured
  • NEXA RECURSOS MINERAIS — 1,649 claims · 17 states · 4.19M ha
  • CBPM, the Bahia state miner — 1,185 claims · 1.35M ha
  • VALE S.A. — 817 claims · 19 states · 1.83M ha
  • CODELCO do Brasil — 518 claims · 1.20M ha
  • Votorantim and CSN Cimentos — 1,191 claims across 15 and 17 states
Why the fragmentation is the opportunity
  • The ten largest holders control only 3.6% of titles; the top hundred, 13.2%
  • That is a market of 55,476 counterparties — no incumbent can see across it
  • Consolidation targets surface by adjacency: who holds the ground beside yours
  • Royalty history attaches to the holder, not only the claim — so you price the operator
Source · ANM Sistema de Cadastro Mineiro (SCM) — titular and CPF/CNPJ per process, joined to the SIGMINE polygon register. Personal tax IDs redacted.
Not a mock-up

This is the live console.

254,715 claims, coloured by commodity, filtered by state, phase and AI score — over satellite basemap, with the deforestation and dropped-area layers on top. Sub-second at national scale.

MineralScan Brazil console
Source · live screenshot of MineralScan Brazil, 254,715 scored concessions, captured 26 Aug 2026.
Satellite confrontation

The title says one thing.
The imagery says whether it is true.

Both boundaries are drawn from our copy of the ANM register, laid over cloud-free Sentinel-2 composites. A register says a polygon exists; imagery says whether anything is happening on it. Held together they separate a producing asset from a speculative land position — the most valuable distinction in this market, and we compute it for all 254,715 concessions.

Sentinel-2 over the Carajas iron concession
Carajás · Parauapebas — PA
  • VALE S.A. · processo 813682/1969 · 30,000 ha
  • The title contains an active open-pit operation — pit benches, haul roads and tailings are plainly inside the boundary
  • Everything around it is closed canopy. The concession is the disturbance
Sentinel-2 over a rare-earth research permit at Moju
Moju — PA · TERRAS RARAS
  • Processo 850171/2023 · 10,000 ha · held by an agro-industrial company, not a miner
  • A perfect 10 × 10 km research square over undisturbed forest — nothing has been dug
  • Plantation blocks press against its eastern edge. This is a position, not a mine
Sources · Copernicus Sentinel-2 L2A surface reflectance via Google Earth Engine — cloud-masked median composites, Jun 2025 – Aug 2026 (67 scenes Carajás, 11 scenes Moju) · boundaries from the ANM SIGMINE cadastral register held in mineralscan_gold. Rendered 26 Aug 2026.
Proof it travels

Brazil was the hard one.
Then we did eight more.

Each country has its own agency, its own formats and its own language. The pipeline absorbs that; the schema does not change.

CountryRecordsCountryRecords
🇧🇷 Brazil18,505,800🇲🇽 Mexico194,426
🇵🇾 Paraguay2,098,605🇻🇪 Venezuela183,418
🇱🇷 Liberia1,878,196🇩🇴 Dominican Republic88,106
🇵🇪 Peru383,199🇧🇴 Bolivia255,829
🇨🇱 Chile354,079Total23,941,658
Why this matters to a Canadian operator

A mineral company working Latin America has to deal with nine regulators, nine data standards and nine languages. We have already normalised all of them into one queryable surface. Adding Canada is a pipeline configuration, not a rebuild.

Sources · national mining authorities — ANM (BR) · SGM (MX) · MOPC/Viceministerio de Minas (PY) · MinPetromin (VE) · MME/LEITI (LR) · DGM (DO) · INGEMMET (PE) · SERGEOMIN (BO) · SERNAGEOMIN (CL).
The same engine, other ground

Minerals were where we started.

Every platform below runs on the identical ingestion, scoring and visualisation stack. Different registry, same machine.

LogisticScan41.6M records · roads, rail, ports, corridors
FloodScan38.3M · flood & wetland risk, USA
DisasterScan15.5M · fire, flood, deforestation
AgroScan6.1M · yield, soil, crop stress
Why an operator should care
  • Infrastructure decides whether a deposit is economic — LogisticScan already holds the corridor data
  • Environmental exposure decides whether it is permittable — DisasterScan holds the alert history
  • Water and flood risk decide the mine plan — FloodScan holds the model
The compounding effect

A mineral deposit is never only a mineral question. Every one of these layers is a variable in the same investment decision — and we are the only party that holds all of them in one schema, for the same ground. That is not a product line. That is a moat.

Source · NumBird production warehouse — per-platform record counts measured on Google BigQuery, 26 Aug 2026.
Case in production · United States

FloodScan: the same engine,
pointed at real estate.

Property is priced on location, condition and comparables. Almost nobody prices the water. We scored every parcel in the state of Florida for flood exposure, then extended the model nationally.

10,864,584
parcels scored
$88.7B
historic claims analysed
2,721,780
flood claims · NFIP
0.771
model AUC · severe loss
Who it is for
  • Real estate — buy, hold or avoid, parcel by parcel, before the offer
  • Insurance & reinsurance — exposure priced on 2.7 million real claims
  • Lenders — collateral risk over the life of the mortgage, not at origination
  • Disaster response — where to stage before the event, not after
Why it belongs in this deck

FloodScan shares every component with MineralScan — the same ingestion, the same Earth-observation layer, the same scoring engine, the same console. Only the registry changed. That is the proof that this is a platform and not a product: 38.3 million records in a completely different industry, built on the machine we made for minerals.

The transferable claim

If a Canadian operator asks whether we can score their ground on their variables — we have already done it twice, in two industries, on two continents.

Sources · FEMA National Flood Insurance Program (NFIP) redacted claims and policies · FEMA National Risk Index · USGS/NOAA elevation and hydrology · Florida county parcel rolls. Model AUC measured on held-out data.
Frontier 3 of 3 · in production

MineralScan × MXchange

Two pieces that close the cycle: one finds and prices the asset, the other turns the asset into capital.

1 · Discovery
  • The full ANM register — 254,715 concessions with title, substance, phase and polygon
  • R$ 74.5 bn of royalty history, so an asset is priced from what it actually produced
  • The holder behind every title — who owns it, what else they hold, in which states
  • Satellite confrontation: the polygon versus what the imagery shows
2 · Liquidity
  • ERC-3643 permissioned tokenisation — compliance enforced at the token, not by a registrar
  • A scored, evidenced asset becomes a tradeable instrument
  • Fractional participation in reserves that are today locked and illiquid
Why the pairing is the point

Tokenising an asset nobody has valued is a spreadsheet with a blockchain attached. The token is only worth what the evidence underneath it is worth — and the evidence is the 127.7 million records. That is why we built the intelligence first and the market second.

Sources · ANM SIGMINE cadastral register and CFEM royalty series (dadosabertos.anm.gov.br) · NumBird scoring models on BigQuery and Vertex AI.
The adjacent market, already legislated

Smart cities are a mandate, not a trend.

Brazil's Lei 14.510 and the national IoT plan (Decreto 9.854) oblige municipalities to instrument and monitor. The EU's Digital Twin programmes and the US Smart Cities initiatives do the same. Every one of them needs exactly what we already run: continuous aerial capture, satellite baselines and automated analysis.

Mapping

Centimetre-grade LiDAR and photogrammetry of the built environment — the base layer every digital twin needs and almost no city actually owns.

Monitoring

Change detection against a 26-year satellite baseline: illegal construction, erosion, drainage failure, vegetation encroachment.

Decision

The same scoring engine that ranks a mining polygon ranks a road segment, a flood-prone block or a slope at risk of collapse.

This is not a pivot

DisasterScan already holds 15.5 million alert records and LogisticScan 41.6 million infrastructure records. The smart-city product is a recombination of systems already in production — which is why we can stand it up in weeks rather than years.

Sources · Brazil Lei 14.510/2022 and Decreto 9.854/2019 (Plano Nacional de IoT) · European Commission Destination Earth / Local Digital Twins · US DOT Smart City programmes.
The disruption, said without hedging

Nobody is closing the whole loop.

CapabilityDrone surveyWingtra · Delair · PerceptoSatellite analyticsPlanet · Descartes · SatellogicAI explorationKoBold Metals · Earth AIMining data & advisoryS&P Global · Wood Mackenzie · SRKNUMBIRD
Autonomous aerial captureYes———Yes
Multi-decade satellite baseline—YesPartial—Yes
Domain-scored predictions—PartialYesYesYes
Full national title & royalty register———PartialYes
Model tasks the next collection——Partial—Yes
Asset → tradeable instrument————Yes
Why the margin is structural
  • Marginal cost of the next country is pipeline configuration, not headcount
  • Marginal cost of the next customer on an existing country is near zero
  • The models improve with every flight — the asset appreciates as it is used
Why it is defensible
  • Nine regulators already normalised — that work is years, not months
  • A 26-year baseline cannot be bought retroactively
  • Built by three USCIS-examined founders with no external capital
Basis · capability comparison against the publicly documented offerings of the named vendors — company websites, product documentation and public filings, Aug 2026. Categories are representative, not exhaustive; each vendor is credited where its public offering delivers the capability.
The prize

$14 trillion, under-mapped.

That is the estimated value of undiscovered mineral resources still in the ground. The constraint on reaching it has never been drilling capacity. It is knowing where to point.

Exploration spend

Roughly $13 bn a year globally, and the industry's own studies put the discovery hit-rate in the low single digits. Most of that money buys information, not ore.

Where we sit

We do not compete for drilling budget. We reallocate it — moving spend from low-ranked ground to high-ranked ground before a rig is mobilised.

The adjacent pools

Infrastructure, agriculture, disaster response and municipal monitoring are each larger markets than mineral exploration — and we already hold production systems in all four.

The honest framing

We are not claiming to have found $14 trillion. We are claiming that the decision of where to look is currently made with a fraction of the available evidence — and that we have already assembled that evidence for nine countries.

Sources · US Geological Survey Mineral Commodity Summaries 2026 · S&P Global Market Intelligence world exploration trends.
What happens next

Open the console. Pick a target.

We are not asking for a commitment to a structure that does not exist yet. We are offering a proof of concept on ground you care about.

StepWhat it isWhat you get
1 · Live demonstrationThis week The production consoles, on real data, driven by you — not a recorded walkthrough.
2 · Proof of conceptOne area of interest Your ground, scored: satellite baseline, regulatory position, infrastructure access, environmental exposure and a ranked target list.
3 · Flight campaignWhere the POC justifies it Autonomous LiDAR and hyperspectral capture over the highest-ranked polygons, feeding straight back into the model.
4 · Standing platformOngoing Your own instance — your entity, your assets, your team's logins, updating daily.
What we need from you
  • One area of interest, however small
  • Whatever historical data you already hold on it
  • One technical person who can tell us when we are wrong
What we do not need
  • No investment
  • No exclusivity
  • No change to how you work today

Above expectations.
Beyond possibilities.

127.7 million records · 1,247 tables · 29 models in production · 10 countries live. All of it can be opened in a browser, today.

🇺🇸 United States🇧🇷 Brazil🇵🇾 Paraguay🇵🇪 Peru🇨🇱 Chile 🇧🇴 Bolivia🇻🇪 Venezuela🇩🇴 Dominican Rep. 🇲🇽 Mexico🇱🇷 Liberia

thenumbird.com · NUMBIRD · MARKUS IT LLC · Florida, USA
Confidential — prepared for our partners of BRISA America

Source · NumBird production warehouse, Google BigQuery — measured 26 Aug 2026.