Our technology
How our technology works, and how it stays secure.
Our software reads the warnings a mine already collects, ranks them by what they put at risk, and tracks each one until the work is done. It advises people. It never controls equipment.
Where AI earns its place
We use AI where it does work a person cannot do at scale, and plain software everywhere else.
| Task | What the AI does | What stays plain software |
|---|---|---|
| Reading condition-monitoring reports | Language models extract the asset, fault mode, severity and recommended action from PDFs and free-text work orders. | Storing the finding, its due date and who owns it |
| Matching tags to assets | Fuzzy matching links names such as "Pump 4 DE bearing" to the right entry in the asset register. | The asset register itself, which stays in the mine's CMMS |
| Checking a warning against plant data | Anomaly detection on historian tags (current, temperature, vibration) shows whether a trend supports the finding. | Thresholds, alarms and trend charts |
| Writing the weekly summary | A draft summary of open risks, which a planner edits and approves. | Overdue lists, closure rates and the rand calculation |
- Decision support onlyNo model writes to a PLC, SCADA setpoint or permit system.
- Every output shows its evidenceThe page, trend or work order it came from.
- A person approvesNothing becomes a work order without a named approver.
- We publish our missesFalse-alarm rates and lead times are reported, not hidden.
- Data stays where the mine allowsOn-site or South African hosting. We send no plant data to a public AI service without written consent.
Security and OT
Built to pass a mine's IT and OT security review.
A mine is not an office network, and a cyberattack can stop a plant. Our design keeps us out of the control layer entirely.
Hosting
N³ serviceSouth African cloud region, or a server at the mine
↑ outbound HTTPS only
Levels 4–5
Business ITCMMS, email, reports
DMZ
N³ site agentReads exports and a historian replica. Accepts no inbound connections.
Level 3
Site operationsHistorian, alarm logs
Levels 0–2
ControlPLCs, SCADA, field devices
- Read-onlyWe read copies and exports. We never write to plant systems.
- No inbound accessThe site agent connects out. No vendor remote sessions are needed.
- Out of the control layerNothing we install touches PLCs, SCADA or field devices.
- Data sovereigntyHosting in South Africa or on the mine's own server, with the same software.
- POPIA by designNames in work orders are pseudonymised on arrival. Access is by role.
- Works offlineThe agent buffers data and syncs when the link returns.
What the research taught us
Technology wins in mining when it starts with a measured problem on site.
We studied 14 companies and projects, from Komatsu's autonomous haulage to platforms that spent billions before proving a single use case. These rules come from that work.
- Problem first, product laterWe find and measure problems before we build anything.Lesson from GE Predix and Argo AI
- Mining people in the roomWe work alongside each mine's own staff and experienced mining practitioners.Lesson from Plotlogic and KoBold Metals
- One use case, proven, before a platformWe recommend one fix at a time, proven against a measured baseline.Lesson from GE Predix and Komatsu FrontRunner
- No AI claims we cannot showEvery figure we report traces back to a data source or an interview at the mine.Lesson from Builder.ai and IBM Watson Health
- Plan for integration and peopleLevel 9 shows that technology alone does not change outcomes.Lesson from Rio Tinto AutoHaul
- Work with what the mine already hasAny fix must connect to the radios, networks and systems in use.Lesson from Newtrax
- Earn trust on the floorWe brief organised labour and health and safety representatives before we start.Lesson from Nautilus Minerals
What our technology will never do
- Control equipment, or write to PLCs and SCADA
- Take the place of certified safety systems such as collision avoidance
- Replace a mine's CMMS, historian or fleet system
- Grow into a large platform before one module has proven its value
- Use AI where a spreadsheet would do just as well
- Make a claim we cannot show with the mine's own data
Next step
Ask us to walk your IT and OT teams through the design.
We are happy to answer a mine's security questionnaire before any data is shared.