Mining & Metals 4.0 · AI and digital integration, pit to port
Rareore.ai integrates AI, industrial IoT and a site-wide digital twin across exploration, extraction, processing and logistics. Decisions move faster, cost per tonne falls, and people step out of the line of fire.
Open pit and underground · Copper, iron ore, coal, gold, bauxite
Illustrative telemetry from a demo site
For the COO
We price every initiative against your cost per tonne over the full asset life: hardware, integration, licences, change management and the people to run it. If a use case does not improve TCO against a measured baseline, we tell you before you fund it.
Autonomy, remote operation and predictive insight remove people from hazardous zones. Safety-critical functions run locally and fail safe, so no protective system ever waits on a network or a model.
| Value lever | KPI we move | Measured against |
|---|---|---|
| Autonomous haulage | Cost per tonne moved, truck utilisation | 12-month pre-pilot baseline |
| Drill and blast optimisation | Fragmentation P80, downstream mill throughput | Blast-by-blast reconciliation |
| Process control with machine learning | kWh per tonne, reagent cost per tonne, recovery | Plant historian, A/B by shift |
| Predictive maintenance | Unplanned downtime hours, mean time between failures | CMMS work-order history |
| Remote operation | Exposure hours in hazardous zones | Personnel tracking and roster data |
| Carbon monitoring | tCO₂e per tonne of product | Metered fuel and power, auditable |
| Cloud cost optimisation | Cloud cost per tonne, idle capacity | AWS billing data tagged by site |
End-to-end value chain
Each stage writes to the digital twin and reads what the others know, so a decision in the pit is priced in the plant and at the port.
Digital twin
Geology, fleet, plant and logistics sit in one spatial, time-synced model. Planners test a blast before it is drilled. Controllers see a wall move before it fails. Finance sees the cost of a decision while it can still be changed.
Platform
Three technology pillars carry the value chain: compute at the edge, language models grounded in your own records, and vision that measures the physical process continuously.
Edge computing
Inference runs where the data is created and keeps running when the network drops. Safety functions never wait on the cloud.
Store and forward · OPC UA · MQTT Sparkplug BRead the pillar →Generative AI for engineers
Geological reports, shift logs and work orders, searchable in plain language, with the source page cited for every answer.
Grounded retrieval · role-based access · on-premise optionRead the pillar →Computer vision
Cameras over the conveyor size every particle and close the loop from blast design to mill throughput.
P20 / P50 / P80 · oversize · belt conditionRead the pillar →IT services · AWS cloud
Rareore.ai is an AWS Select Tier Services Partner. We design, deploy and manage the AWS environments mining operations depend on, so one partner is accountable from the sensor on the drill to the monthly cloud bill.
Consult
Readiness and cloud TCO assessment, migration roadmap and target architecture.
AWS cloud services →Deploy
Landing zone, workload migration, infrastructure as code and hybrid deployment at site.
AWS cloud services →Run
24×7 monitoring and incident response, security operations, backup, disaster recovery and FinOps.
AWS cloud services →Build
Industrial data lake, IoT ingestion, digital twin and AI platforms on AWS.
AWS cloud services →How we engage
Each phase ends with a measured result and a decision point, so capital follows evidence.
TCO baseline, data and OT readiness audit, and a ranked list of value levers with expected payback.
One lever on one site, measured against the baseline with a success threshold agreed before work starts.
Site-wide twin, integration with ERP, MES, CMMS and OEM systems, and operator training.
Managed AWS cloud operations, analytics for the remote operating centre, model monitoring and continuous improvement.
Book a TCO assessment
In four to six weeks we build a cost-per-tonne baseline for one operation and rank the levers worth funding. The baseline model is yours to keep.
Technology pillar · Edge computing
Underground, connectivity drops at every new heading, GPS does not exist, and heat and dust shorten the life of anything exposed. We design for those conditions first, then connect to the cloud when the network allows.
Collision avoidance, gas trend detection, ventilation on demand and equipment health models run on ruggedised nodes on each level.
Every reading is time-stamped and buffered at the edge, then synchronised in order when a link returns. Nothing is lost and nothing arrives out of sequence.
Protective functions never depend on the cloud or a model. If the network fails, interlocks and alarms keep working on site.
OPC UA and MQTT Sparkplug B from sensor to cloud, so your data is never locked to one equipment vendor.
Tag-based and inertial tracking for people and machines, fused at the edge and written to the twin with the rest of the telemetry.
Models are signed, versioned and rolled out when bandwidth allows, with automatic rollback if a new version underperforms.
Interactive
Switch the in-mine link between level 14 and the portal gateway. Local inference and safety interlocks continue; data waits in the edge buffer and backfills in order when the link returns.
| Layer | Where it sits | What runs there | Connectivity |
|---|---|---|---|
| L0–1 Sensors and machines | Face, drives, plant floor | Gas, vibration, ventilation, LHD and truck telemetry, personnel tags | Wired fieldbus, Wi-Fi, BLE tags |
| L2 Edge node | Each underground level or pit sector | Inference, filtering, buffering, local alarms and interlocks | Leaky feeder, Wi-Fi 6, private LTE/5G |
| L3 Site gateway | Portal or site data centre | Historian, site twin, OT security boundary, aggregation | Fibre backbone |
| L4 Cloud and ROC | Regional ROC and cloud region | Fleet-wide analytics, model training, enterprise integration | Redundant WAN, satellite backup |
Technology pillar · Generative AI for engineers
Most mines hold decades of knowledge in drill logs, geological reports, shift notes and maintenance records that nobody has time to read. We build language models grounded in that archive, so engineers get answers with the evidence attached.
Interactive
Pick a question. The assistant retrieves the relevant pages and answers only from them, with each claim cited. Sample corpus of fictional documents.
Sources cited
Scanned reports, handwritten logs, CMMS exports, historians and incident files, with OCR tuned for mining forms.
Drillholes, assets, benches, dates and people are extracted and linked to the twin's coordinates and asset tree.
Retrieval over a governed index, with a model adapted to mining vocabulary and units.
Every claim cites its page. When the archive holds no evidence, the assistant says so.
Access mirrors document permissions; every question and answer is logged for audit.
Maintenance
A planner asks why an asset keeps tripping and gets the work orders, root-cause analyses and the interval that changed the outcome.
Safety
Before a task starts, the crew sees past incidents on the same equipment and the controls that followed. Incident learning reaches the face.
Geology
Old drilling campaigns become searchable by location and depth, with their QA limits stated, so targets are not drilled twice.
Technology pillar · Computer vision
Fragmentation decides how fast the crusher and mill can run. Cameras above the conveyor size every particle continuously, so the result of each blast is known within minutes and the next blast is designed with evidence.
Interactive
Switch the blast design and watch the size distribution on the belt change. Sizes are reported by mass, as a sieve would.
P20, P50 and P80 by truckload or time window, mapped back to the blast pattern that produced the rock.
Boulders and stray metal are flagged before they block a crusher or rip a belt.
Edge wear, splice damage and misalignment are detected while the belt runs, so repairs are planned rather than forced.
Results feed the drill-and-blast model, the twin and the mill controller, so each blast improves on the last.
Charge per hole set from measure-while-drilling hardness.
Belt cameras report the size distribution within minutes of loading.
Crusher gap and SAG mill feed rate respond to the incoming rock.
Throughput and kWh/t are reconciled against the pattern and fed into the next design.
| Item | Specification |
|---|---|
| Imaging | Stereo cameras with strobe lighting, sealed housings rated for dust and vibration |
| Processing | Edge GPU at each transfer point; results streamed to the twin, raw frames retained on demand |
| Calibration | Against sieve samples at commissioning and on a fixed schedule thereafter |
| Outputs | P-values, oversize events, belt alarms; OPC UA tags for the plant control system |
IT services · AWS cloud
As an AWS Select Tier Services Partner, Rareore.ai designs, deploys and manages the AWS environments that mining operations depend on, from the landing zone to the incident call at 2 a.m. The team that builds your digital twin also runs the cloud underneath it, so one partner is accountable from sensor to cloud bill.
The CIO question
Who is accountable when the historian, the twin or the ERP goes down at shift change?
Mines often split infrastructure, applications and analytics across several vendors, and incidents fall between them. We take the whole stack on one contract, with one team accountable for every incident.
Consult
Cloud readiness and TCO assessment, a migration plan ranked by value and risk, landing-zone and security architecture, and AWS Well-Architected reviews of workloads you already run.
Deliverables · cloud TCO model · migration roadmap · target architecture
Deploy
A multi-account landing zone, migration of ERP, mine-planning, historian and GIS workloads, infrastructure as code with CI/CD, and hybrid deployment at site on edge nodes or AWS Outposts.
Control Tower · Application Migration Service · Database Migration Service · DataSync · Outposts
Run
Monitoring and incident response around the clock, patching, backup and disaster recovery, security operations, and a monthly FinOps review with cost per tonne on the first page.
CloudWatch · Systems Manager · AWS Backup · Elastic Disaster Recovery · Security Hub · GuardDuty
Build
The industrial data lake, IoT ingestion from every site, the digital twin, and the machine-learning and generative-AI platforms behind each technology pillar.
IoT Core · IoT Greengrass · IoT SiteWise · IoT TwinMaker · S3 · Lake Formation · SageMaker AI · Bedrock
| Layer | What runs there | AWS services |
|---|---|---|
| Site edge | Inference, buffering and local control; OT networks stay isolated | IoT Greengrass on edge nodes; Outposts at large sites |
| Connectivity | Encrypted links from each site to the region, with a backup path | Direct Connect, Site-to-Site VPN; satellite backup at remote sites |
| Landing zone | Separate accounts for OT data, IT, analytics, security and logs | Control Tower, Organizations, IAM Identity Center |
| Data | Industrial data lake with governed access | S3, Lake Formation, Glue, Redshift, IoT SiteWise |
| AI | Model training, deployment to the edge, grounded generative AI | SageMaker AI; Bedrock with Knowledge Bases |
| Security and resilience | Encryption, threat detection, audit, backup and disaster recovery | KMS, GuardDuty, Security Hub, CloudTrail, AWS Backup, Elastic Disaster Recovery |
| Regions | Indian clients: data residency, with disaster recovery between two regions. Global clients: the region nearest the operation | Asia Pacific (Mumbai), Asia Pacific (Hyderabad) |
Interactive
Cloud is part of the TCO you are measured on. We tag every resource by site, stage and cost centre, so cloud spend is reported per tonne alongside haulage and power. Try it with your own figures.
Example inputs · replace with your own billing and production figures
Each tier sets what we monitor, how we protect it, and how often we review its cost.
| Tier | Essentials | Advanced | Mission-critical |
|---|---|---|---|
| Suits | Corporate IT, non-production | ERP, mine planning, GIS | Historian, digital twin, ROC systems |
| Monitoring | Business hours, client time zone | 24×7 | 24×7 with a named engineer |
| Backup and DR | Daily backup, annual DR test | Daily backup, DR test twice a year | Cross-region DR, quarterly test |
| Security | Monthly posture report | Continuous threat detection | Continuous detection, OT network monitoring, CERT-In runbook |
| FinOps | Quarterly review | Monthly review | Monthly review with cost per tonne by site |
For PSUs and other regulated operators. Global clients get the same controls, mapped to local law and the AWS region nearest the operation.
Workloads and backups stay in the AWS Asia Pacific (Mumbai) and (Hyderabad) regions, with disaster recovery between the two.
Logging and incident runbooks are set up for CERT-In's 2022 directions: incidents reported within six hours, and system logs kept in India for 180 days.
Personnel tracking and HR data are handled under the Digital Personal Data Protection Act, 2023, with access limited by role.
Mine and plant networks are segmented from corporate IT using IEC 62443 zones and conduits. OT data reaches the cloud through one controlled path.
Every change and every access is logged centrally, in a separate security account that operations staff cannot alter.
Where a workload must stay at the mine, AWS Outposts or edge nodes run it locally under the same controls.
An AWS Partner at Select tier on the services path.
Recognised by the Department for Promotion of Industry and Internal Trade.
Registered under the Ministry of Micro, Small and Medium Enterprises.
Team AWS certifications by name and count, and Well-Architected reviews completed.
Workload inventory, cloud TCO model, compliance review and a migration plan in waves.
Landing zone, network links to each site, identity, security tooling and cost tagging.
Workloads move in waves, each with a tested rollback and a cutover agreed with operations.
Managed operations at the agreed tier, with monthly service and cost reviews.
Where to start
We review one workload against the AWS Well-Architected Framework and report the risks, the fixes, and what it costs to run per tonne.