“The environment is just a parameter”: Sohan of Edgegenix on bushfire detection, aged care robots, and one edge AI stack

Ecosystem Profile, Global Day Zero. Issue 13, week ending 13 September 2026.

Most computer vision still phones home to the cloud. Edgegenix runs it on a device at the powerline, the mine site and the fire ground, where there is no connectivity to phone home with. The same stack also sits inside a companion robot in an aged care facility, which sounds like two companies until you hear how Sohan describes it.

Give us your 15 second intro and what you’re working on.

I’m Sohan from Edgegenix. Most AI vision still needs the cloud. We put it on the edge, on a device at the powerline, the mine site, the fire ground, so it works where there’s no connectivity. Wildfire detection, vegetation encroachment, PPE compliance, asset inspection.

Two things.

The first is the core platform, deploying computer vision models at the edge for wildfire detection, vegetation encroachment along power corridors, PPE and exclusion-zone compliance on mine sites, and roof and powerline inspection. Right now that means turning pilots into signed deals with councils, utilities and fire agencies ahead of the summer fire season.

The second is BrightBuddy, a companion robot for aged care, hospitals and homecare. Same edge AI stack, very different room. We’re also opening up the brain behind it, Buddy Intelligence, so anyone with their own robot hardware can build on it.

What were you doing before this, and what finally made you start?

I ran operations for Dryad across ANZ, and before that I headed the innovation team at Fujitsu. Both roles put me close to the problem, but never close enough to fix it properly. I’ve always been entrepreneurial, starting something was a matter of when, not if. AI made it now. The tools finally exist to build products that weren’t possible three years ago, and that window won’t stay open forever.

The gap I kept running into was that nobody had built a good computer vision platform at the edge. Plenty of models, plenty of cloud, nothing that just works on-site. That’s what Edgegenix is: deploy a computer vision model where the problem actually is, wildfire detection, PPE and exclusion-zone compliance, roof and power line inspection.

Are you avoiding AI, somewhat into AI, or all in on AI?

All in, but not the hype version. Everything I build is AI. Computer vision running on edge devices in power corridors, on drones over fire grounds, in aged care facilities. It isn’t a feature I bolted on; it’s the whole product.

What I’m sceptical of is AI as a demo. A model that scores well on a benchmark is easy. The bar my customers set is whether it works offline, on cheap hardware, in bad weather, with nobody babysitting it. That’s a much harder question than “can the model do it,” and it’s where most of the real engineering goes.

What’s the hardest part for startups in the early days?

The gap between interest and a purchase order. Everyone takes the meeting. Everyone finds it interesting. Turning that into a signature, through procurement, budget cycles, and a risk-averse buyer who’s never bought from a company your size, is the actual job, and nothing prepares you for how long it takes.

The second hardest part is sequencing. There are always five good things you could build, and you can only do one of them properly. Choosing which four to say no to, every week, is harder than the building.

Where can people usually find you?

LinkedIn is the fastest way to reach me, I’m responsive there.

Otherwise, Sydney. I’m around the startup and govtech events here, and I’m always happy to grab a coffee if you’re working on anything in computer vision, geospatial, robotics or infrastructure. edgegenix.com and brightbuddy.ai are the two places to see what we’re actually shipping.

What’s one thing people find surprising about you?

That the same company doing bushfire detection, PPE for construction and other computer vision use cases also puts companion robots into aged care homes. It sounds like two businesses that wandered into each other. It isn’t, it’s the same edge AI stack pointed at a different problem. Once you can run vision and inference reliably on a small device in an unpredictable environment, the environment is just a parameter.

People are also usually surprised there’s no big team behind it. Smaller than it looks from the outside.

How can you help others, and what do you need help with?

Happy to help with anything edge AI and computer vision, what genuinely runs on-device versus what still needs the cloud, how to get a v0.1 that’s real rather than a demo, and selling into councils, utilities and government, which is its own sport with its own rules. Also glad to talk candidly about hardware sourcing and OEM partnerships, including the parts that went badly.

What I need: warm intros into vineyards, golf courses, construction companies and mining where there is a need to solve problems using computer vision, and into aged care, home care and hospitals.

Sohan builds edge AI and computer vision at Edgegenix, and companion robots for aged care at BrightBuddy. Reach him on LinkedIn.