Hi Smart Community friends,
We’re continuing our AI in the Real World series, and this episode is a little different. We recorded it in person, and Josh and I interview Michael Cybulski about his journey as a startup founder and entrepreneur.
Michael is the CEO and founder of Retina Visions, who are also the sponsor of this series. Michael, Josh and I talk about how Michael built Retina Visions from a tiny proof of concept, with one camera on one ute at Moreton Bay Council, into a company now working with over 50 councils across Australia.
We discuss how Retina Visions uses computer vision, and more recently stereo vision cameras, to automatically detect things like potholes and cracking from council vehicles. This replaces manual road inspections that hadn’t really changed in the last 40 years or so. We also talk about why AI is such a broad term, the realities of building and rebuilding a product as the technology changes underneath you, and the genuine safety benefits of taking inspectors off the side of busy roads.
Along the way, Michael shares what he’d do differently (including a very expensive lesson about building your own camera), his “fail fast” approach to testing ideas, and what being a CEO has really come down to for him. I also reveal how Michael’s data ended up in my Master of Data Science project.
As always, we hope you enjoy listening to this episode as much as we enjoyed making it.
What we cover in this episode:
- How Michael went from a QUT asset management project to founding Retina Visions in 2017
- The idea that started it all, including its unlikely link to Grand Theft Auto
- Why road inspections had barely changed in 40 years, and what inspectors do now
- How Moreton Bay Council and Arup helped turn a concept into a pilot with one camera on one ute
- Navigating two years of trials and proofs of concept before landing a rollout contract
- Building cameras with Raspberry Pis, moving to mobile phones, and now to stereo vision with depth data
- The challenge with ‘AI’ being such a broad term
- The reason “agentic AI” isn’t the interesting part of Retina Visions’ pipeline
- The safety benefits of detecting defects without stopping on high-speed roads
- How measurable data makes severity scoring more consistent and helps councils triage
- The “virtual road condition score” and using Retina Visions data for capital works planning
- Michael’s advice for founders: move fast, persist, and learn from failure
- What’s next: a continuously updating digital twin for councils
Quote options:
“Now, because we’re able to measure the defects, our severity scores are way more consistent because it’s based on measurement data. And that allows the council to triage and basically say, ‘I only want to see defects that hit this threshold.’” ~ Michael
“You shouldn’t be too married to any idea when technology can change that quickly, that maybe an idea that was good a month ago isn’t good anymore.” ~ Josh
“Safety is definitely [an] element that councils take up the [Retina Visions] technology for. Getting crews to be able to stay in their vehicles or just stay in the office and triage defects that are being picked up by their waste fleet.” ~ Michael
“We’re not replacing the engineer here … their judgment is still needed. And now they have a wealth of data to deal with.” ~ Zoe
“[A regional Council] will have tens of staff, less than 100 staff, and a network of 3,000 km to look [after]. … This is something that helps them.” ~ Michael
Links:
- Retina Visions
- Michael Cybulski on LinkedIn
- Michael’s previous episode https://zoeeather.com/scp-e31-a-tailored-approach-to-tech-with-michael-cybulski/
Connect:
Connect with me via email: hello@mysmart.community
Connect with My Smart Community via LinkedIn or YouTube.
The Smart Community Podcast is produced by Ellen Ronalds Keene from PERK Digital
