Edgegenix puts AI at the edge — where your operation actually happens. Edge AI runs purpose-built detection models on-device, beside the camera, even with no network. Cloud AI verifies every alert and keeps a timestamped evidence trail — so detection and evidence never depend on the network being up.
Fire was first. But the platform doesn’t stop at one hazard — a new use case is a new model, not a new product. Choose your industry below to see how Edgegenix adapts to your field. Every card follows the same pattern: the problem, the decisions that matter, and how Edge AI and Cloud AI combine on the cameras you already operate.
Verified smoke and fire detection running in production today — on real camera streams, on real hardware. Built for the properties and operations below.
Players, staff and guests spread across eighteen holes, often backing onto scrub and far from shelter. On high-risk days the club decides whether to open, when to clear the course, and how to show it met its duty of care.
One camera on high ground watches the full property and the bushland beyond it, day and night.
Minutes of notice with location and direction turn evacuation into a marshalled clearance, not a scramble.
Timestamped warnings and actions — the documented watch for the board, members and insurers.
Confidence to make open and close calls on total fire ban days, with a verified watch running.
Smoke & flame detection on site — keeps watching with no connectivity, syncs when the network returns.
Verified alert with location and direction, fused with weather — plus the duty-of-care evidence trail.
a fire's area can grow within two hours of ignition — every ten-minute delay adds up to 1,500m² in moderately dry conditions.
Smoke taint can destroy a vintage while flame never reaches the rows — and claims turn on proving when grapes were exposed. The decisions that matter are when to sample, when to pick, and what to show the insurer.
A verified alert while smoke is still distant, so sampling and picking calls can start early — aligned with AWRI sampling protocol.
A timestamped, directional record of when smoke arrived, how long it stayed and from where.
Bring picking forward with confidence when smoke risk rises during vintage.
A continuous watch on nearby burns — direction, behaviour and proximity, day and night.
Smoke detection over the rows and the boundary — on-device, day and night.
Exposure timeline fused with wind and terrain — the evidence that targets sampling and resolves a contested claim.
estimated cost to Australian wine grapes from the 2020 bushfires — largely smoke taint, not flame.
Huge areas, few people, patchy connectivity — and detection that still leans on lookout towers and public calls. In plantation timber, a late detection is the difference between a contained ignition and a lost compartment.
Camera detections verified at the edge before escalation, so crews aren't sent to chase dust and glare.
Perimeters, hotspots and containment tracked through planned burns, with a record of conditions.
Boundaries, direction and intensity from thermal and RGB imagery, fused with weather and terrain.
Burn severity mapping, salvage priorities and regeneration monitoring after an event.
Verification at the tower — smoke vs dust, cloud and glare — before a crew moves.
One estate-wide picture: detections, burns, assets and compliance records.
of NSW timber plantations were destroyed in the 2019–20 Black Summer, with more than 70% hit by severe canopy fire.
Schools near the bushland interface carry the sharpest duty of care there is: students spread across grounds, buses and pickups on a schedule, and closure calls that have to be made early — on information, not rumour.
An AI watch on the boundary and beyond it — a verified alert with location and direction, not a smell of smoke and a guess.
Minutes of notice turn a bell and a hope into a marshalled movement to the assembly area or buses.
Evidence for the early calls — closures, excursion cancellations, early pickups — before the situation forces them.
A timestamped record of watch, warning and response for the department, board and parents.
Continuous smoke & fire watch on the school boundary — works through network congestion on emergency days.
Verified alerts to leadership with direction and context — and the record behind every call.
of verified notice are the difference between an orderly, marshalled movement of students and an improvised one.
Residents with limited mobility can't be evacuated in minutes — a safe relocation takes staff, vehicles and time. The earlier the verified warning, the more of that time you have; the record of it protects residents, staff and the provider.
Verified detection while fire is still distant — the hours a staged, safe relocation actually needs.
Location and direction context to decide: shelter in place, prepare, or move — wing by wing.
A timestamped record of watch, warning and action for regulators, insurers and families.
Early notice of smoke arrival for residents with respiratory vulnerability — not just flame risk.
24/7 AI watch on the boundary and the interface — independent of staff eyes on a busy day.
Verified alerts with spread direction — decision support for shelter vs relocate, and the evidence trail.
a safe relocation of low-mobility residents takes hours, not minutes — early verified warning is the only way to buy them.
Substations, water treatment plants, comms towers, depots — critical assets sit in bushfire-prone country with nobody on site. A fire that starts near one is found late, and the asset owner answers for the watch that wasn't there.
Continuous AI detection at unmanned sites — smoke and flame around the asset and the approach to it.
Verified alerts with time to act — de-energise, dispatch, or activate suppression before the front arrives.
Detection keeps running when the link drops, and alerts catch up the moment it returns — no gap in the watch.
The documented watch and response for regulators and insurers — per site, timestamped.
Detection on the asset — rugged, pole-mounted, offline-capable. No control room required on site.
Fleet of sites on one operating picture — alerts, spread context and the audit trail.
most interface assets are unmanned — an AI watch is the only continuous watch they will ever have.
The Edge AI Core that watches for smoke is a reusable foundation — a new use case is a new detection model, deployed over the air to the same hardware, not a new product. These are the operations we’re building toward with partners.
Powerline failures caused the fires responsible for most of Black Saturday's 173 deaths — an ageing line felled in wind, a corroded tie wire, a branch grown into a conductor. Inspection is periodic; the network degrades continuously. Edge AI on inspection drones and fixed points turns every flight into an inspection pass, so what changed since last time gets seen before it arcs.
Detection models on inspection drones and fixed points — every flight becomes an inspection pass, processed on the drone itself.
Growth approaching conductors flagged with GPS-tagged imagery, prioritised by clearance distance.
Hot joints, failing components and arcing risk spotted by thermal imaging before failure — and before ignition.
Every finding timestamped, located and evidenced — the inspection record regulators and insurers expect.
Encroachment and thermal-anomaly models on-device — on the drone or on the pole, network or not.
Corridor-wide picture: findings ranked by risk, work orders raised, the audit trail kept.
the settlement over the Black Saturday Kilmore East fire — started by a failed powerline — Australia's largest class action settlement.
Autonomous haulage safety logic does the only thing it's allowed to when it sees an object it can't classify: stop the truck. Kangaroos are the worst case — erratic, fast, and famous for confusing even road-going autonomy. Every unnecessary stop ripples through the circuit; fleets budget hundreds of hours a year of stoppage downtime. Purpose-built wildlife models classify the object and its trajectory in milliseconds — so the fleet keeps rolling for a roo that's clearing the haul road, and still stops dead for a person. Every time.
Kangaroo, vehicle, person — classified on-vehicle in milliseconds, no round trip to a server the pit can't reach.
A passing animal predicted to clear is advisory context, not an emergency stop — productivity gains without touching the safety case.
The safety hierarchy is preserved absolutely: people and vehicles trigger full response, wildlife triggers judgement.
Detections logged with footage and telemetry — the record that tunes the models and supports the safety case.
Wildlife and object models on the vehicle — rugged, offline-capable, updated over the air.
Fleet-wide encounter picture — hotspots, patterns, and the evidence trail for the safety case.
of stoppage downtime budgeted per autonomous fleet per year — millions of dollars in lost opportunity from stops that didn't need to happen.
A utility-scale solar farm is hundreds of thousands of modules spread over square kilometres, and a wind farm is turbines nobody can reach without a rope team or a shutdown. Faults do not announce themselves — a cracked cell, a hotspot from a failed bypass diode, a soiled or shaded string, erosion on a blade's leading edge. They quietly take production away until someone walks the rows or climbs the tower. Flying the asset with visual and thermal cameras turns that into a scheduled pass, with the detection running on the drone rather than in an office weeks later.
Cracked cells, thermal hotspots, failed bypass diodes, soiling and shading flagged per module — visual and thermal, processed on the aircraft.
Every finding carries string and module identity with GPS, so a crew walks to the fault instead of looking for it.
Leading-edge erosion, surface cracking, lightning-strike damage and delamination flagged from the flight, without a rope team on the tower.
Each pass is timestamped and kept, so this quarter can be put beside last quarter and degradation seen as a trend rather than a surprise.
Defect models running on the drone — a fault is flagged during the flight, not after the footage is uploaded.
Every finding ranked by production impact, mapped across the site, and kept as the inspection record.
A pass over an array or a turbine replaces the walk-the-rows inspection and the rope-access climb — the same cameras, the model doing the looking.
Australian farms recorded 74 deaths in 2024 — a twenty-year high — and farm work is lone work: early starts, machinery, livestock, no one in earshot. The difference between an incident and a fatality is often how long until someone notices. A camera in the shed becomes a safety observer: the on-device model recognises a person down and not moving, and raises a verified alert with footage — no wearable to forget, no button to reach.
A person down and not moving recognised on-device — alert raised in seconds with footage to verify at a glance.
The watch is on the shed, not the worker — nothing to charge, forget, or fail to reach after a fall.
Rural links drop; the watch doesn't. Detection runs on-device and alerts sync the moment the link returns.
The same camera platform extends to stock monitoring, shed operations and fire watch over the paddocks.
Man-down model on the shed camera — on-device, day and night, independent of the internet.
Verified alert escalation — family, manager, or emergency contact, with the evidence attached.
work-related deaths on Australian farms in 2024 — the highest in twenty years. Agriculture is Australia's second most dangerous sector.
The same Edge AI Core that watches for smoke today is the foundation for all three — encroaching branches, wildlife on haul roads, a worker down in a shed. Each is a new model on the same platform, not a new product.
A structured pilot on your site — clear milestones, measurable results, no obligation. We validate detection, alerting and the evidence record against your operational requirements, on your existing hardware or ours.