Veheri - AI-Assisted Teleradiology Platform

Who is Oz and what’s his story?

Oz started as an engineer in India but found coding without a soul about as satisfying as a soggy burrito. A psych class flipped the switch, leading to a degree in computer science and an obsession with Human-Computer Interaction (HCI). From AR apps to game interactions, Spain became the perfect spot for a master’s in Interaction Design (and better tapas). Since then, he’s built brand installations and spent six years in medical tech, turning complex systems into intuitive experiences.

Why Oz, when there are 10,000 other designers out there?

Oz doesn’t just make things pretty; he builds UX that works harder than a caffeine-fueled intern. From motion-driven interactions to prototypes that catch problems before they exist, he makes the impossible look easy. His secret weapon? Blending design with nature’s best tricks (shoutout to biomimicry) and using systems thinking to make everything click. Others hand you a shiny UI; Oz makes sure it works. No head-scratching required.

How does Oz tackle projects without losing his mind?

Oz tackles projects like a science experiment, minus the lab coat. Every project follows four phases: Discovery, Concepting, Prototyping, and Refinement. He maps ecosystems, interviews stakeholders, and locks down user needs before diving into rigorous testing. What sets him apart? Collaboration isn’t a checkbox; it’s the core. Feedback isn’t a hurdle; it’s fuel. Design isn’t solo; it’s a team sport, and Oz plays to win.

*Disclaimer

All written by me, in the third person, because the alternative was a paragraph starting with “I am passionate about design”.

A blank product, and freedom the parent company never had.

Collective Minds acquired a small veterinary teleradiology company to enter a new market. I was handed the brand, the principles, and a blank product. Same design system underneath, almost nothing else the same.

Here's how it got built, and why the less-regulated twin became the place we ran our boldest experiments.

The Acquisition

In January 2022, about a year after I joined Collective Minds, my CTO told me we were buying our way into teleradiology. The target was a small veterinary teleradiology company in Germany with good bones and real potential. We would keep its commercial team where it was, build an entirely new engineering team in Barcelona, and I would work with the new CEO to shape the brand and design the product from scratch.

Starting with animals was a deliberate choice, not a smaller ambition. Veterinary medicine is far less regulated than human medicine, which meant we could move fast, experiment in the open, and learn what the market actually wanted before committing. But the products were always meant to be cut from the same cloth as the human-medicine side, so that one day the same foundations could carry human teleradiology too. Two markets, one system, deliberately different speeds.

The name came later and it fit: in ancient times the people who protected animals were called veheri.

Start where the rules let you move. Build so the rules can catch up later.

Same System, Leaner Engine

Veheri runs on the same design system as the Collective Minds products, and that is what got us moving on day one instead of month six. With a team you could count on one hand, inheriting a mature design language meant our energy went into the product instead of into reinventing buttons.

But because it was a leaner company, we built on leaner, faster technology beneath those same components. That let us experiment constantly, ship new features every two weeks, and run a wildcard every quarter where we tried something genuinely untested. Animal medicine isn't regulated the way human medicine is, so we had the room to fail in public and keep only what stuck. Then the most satisfying thing happened in reverse: the faster underlying component tech we built for Veheri was adopted back into the original Collective Minds design system, making the older, larger product faster too. The experiment lab improved the very thing it was spun out of.

The unregulated twin became the R&D lab. What worked there flowed back upstream.

Every Species Is a Different Body

Human radiology has one quiet advantage: every patient is the same species. Veterinary radiology has cats, dogs, horses, birds, and once we entered the Australian market, a long list of animals I had genuinely never heard of. Different species, different modalities, different anatomy, all of it on one platform.

So I designed a request form that adapts to the animal it's about. It reads the patient history with an LLM and helps the vet fill the form quickly while keeping the data accurate, because however relaxed the regulation is, this is still medical data and accuracy is never negotiable. The other structural difference reshaped the entire privacy model: in veterinary medicine the privacy that matters belongs to the pet owner, not the patient. That single fact changes who you are protecting and how, and it ran through every decision downstream.

One platform, every body. The form had to know a horse from a hawk before the vet typed a word.

How Teleradiology Actually Flows

A teleradiology marketplace has two people at its ends, and the whole product is really the bridge between them. After the discovery work, the platform's structure came down to serving these two journeys well.

The vet who needs a report.

A vet somewhere in the world scans an animal, and the scan lands automatically in their institution's library. They pick the case, add a few basic details, and the system fills the rest. They verify it and send it out, and from that first moment they can see the turnaround time they'll get, faster than the competitors they're used to. What comes back is two reports, not one: the clinical report they need, and a plain-language version they can hand to the pet owner.

The expert who writes it.
‍
The case is matched to a relevant expert, who reports using whatever suits them: text, voice, video, or AI assistance, all available, none mandatory. A small transformer learns each expert's writing style and helps them write faster while staying accurate. Behind them sits the operations layer that makes the speed real, experts log their available hours, see which slots run heavy or light so coverage holds 24/7 across every timezone, and timers sit on cases so no one can quietly claim one and disappear while a vet waits.

The clinical report is for the vet. The plain-language one is for the worried person in the waiting room. Both matter.

The Expert Quality System, Not a Game

The obvious word for this is gamification. It is the wrong word, and refusing it is the point. These are medical professionals, and you don't motivate a radiologist with streaks and badges the way a language app would.

What we built instead is a private self-improvement system, visible only to the expert it belongs to. The score is built from the feedback vets give on every single report, quality, detail, accuracy, with an LLM pulling structured insight out of that feedback. It rewards the behaviour that matters: showing up for shifts, writing complete reports, meeting the standard set by the global radiology associations. Doing the job well brings an expert more cases and, in time, more income. To stop the platform tilting toward a handful of high performers, we built a community layer where strong experts teach others, and a path for residents to audit reports and learn first-hand. It is professional development wearing a progress bar, not a game.

Doctors don't need streaks. They need honest feedback, fair work, and a way to get better. That's what we built.

AI That Never Holds the Pen

If the quality system was about trusting the experts, this chapter is about getting them to trust the tool. AI is not trusted in these rooms, and pretending otherwise would have been the fastest way to lose them. So the rule was absolute: AI never overpowers the human, and the entire platform works with every AI feature switched off.

AI is opt-in, something an institution chooses to turn on. When it's on, it speeds the vets and experts up and lifts the quality of the output. It checks for laterality errors in the boxes as an expert writes them. It prompts them not to miss the key clinical questions, based on its analysis of the patient history. Every one of those behaviours was trained with in-house and external vets and radiologists, and tested hard with real users. The human always decides. The AI just makes sure they decide with everything in front of them.

In medicine the question isn't how smart the AI is. It's who's holding the pen. We made sure it was always the human.

Meet Them Where They Work

The same respect-the-professional instinct shaped the viewer. A cloud DICOM viewer is our speciality on the human-medicine side, so every Veheri report shipped with one by default. Then real use taught us that every radiologist on earth already has a viewer they trust, and forcing ours on them was friction, not a feature.

So we integrated with all the major viewers: a case opens in whatever viewer the expert prefers, by direct link or export. Ours stays available for the people who want what it uniquely offers, like queued cases, where an expert clicks through auto-assigned cases one after another without hunting for the next one, and gets a gentle nudge when they hit the number they committed to for the session. Keep going, or call it a day. Their choice, every time.

What the Product Became

Veheri shipped, and it's used hard. Revenue grows quarter over quarter, year over year. I still run regular feedback sessions with both kinds of user, the vets who send cases and the experts who read them, and we're still shipping features every two weeks.

What started as a blank product became a platform with real operational depth. It's a whole operation now, not just the screen where the scan appears: an admin app for case managers, a sub-app for mobile scanning studios working in remote places, a Stripe-based pricing and invoicing workflow, and a chat connecting vets, experts, and admins. It became multilingual by necessity, translated into every language we do business in, because a platform spanning continents can't assume a shared one. And it became genuinely interactive, the web report far richer than the PDFs this industry defaults to, with a feedback visualiser that helps institutions improve their own scanning protocols and external imaging-AI engines that can be plugged in for fully automated instant reports when a case calls for it.

What It Taught Me

The thing I'm happiest about is that I didn't treat Veheri like the human-medicine products. It got to be its own thing. Because the company was so small when we started, it became a playground, for me and the team, to try our wildest ideas, fail quickly, and keep what worked. That's a rare and genuinely fun way to build, and what came out of it helps people every day. In an industry where the speed of a report can decide an outcome, that speed has meant real animals' lives saved because a clinic chose us.

The lesson I carry out of it: a shared system isn't a constraint on a new market. Handled right, the new market becomes the place you discover what the shared system should become next.