Insights from NUManima 2026: AI in Veterinary Medicine

Notes from NUManima 2026 — France’s conference on AI and digital health in
veterinary medicine, organized by association Vet IN Tech.

I went as an outsider of sorts: a biomedical engineer coming from particle accelerators and medical devices, who ended up building a behavioural tracking tool for dogs, sitting in a room of people who’ve spent their careers with animals in a way I haven’t. I went to listen.

I’m glad I did.

The thing I expected wasn’t there

I expected hesitation. Some version of “yes, but” – the cautious professional instinct to wait and see before embracing new technology. It wasn’t there. What I found instead was curiosity. Active, critical, practical curiosity. Vets are testing new tools, evaluating what works, reading the research, and many are building their own digital solutions when nothing on the market fits their needs.

This is not a profession waiting to be dragged into the future.

What they actually need

That said, what they want from AI is specific. Not transformation. Automation of the things that eat their time: answering the phone, generating post-visit reports, reducing the administrative layer that sits between them and the actual work. Give a vet their afternoon back and you’ve done something useful. Abstract promises about reimagining veterinary medicine are harder to get excited about when you have twelve appointments and a waiting room full of anxious owners.
The practical and the visionary are both present in this community, they’re just asking for different things at different moments.

Data and why sharing it isn’t optional

There’s a genuine appetite for data in this room. But the conversation around data is more nuanced than just “more data = better.”
Regulations matter. Personal data of owners, clinical data of patients, the frameworks are still catching up with the tools. And there’s a harder question underneath: if clinical data is siloed in individual practices, progress is slower than it needs to be. Sharing anonymised clinical data is, in the blunt framing I heard, not just useful – it’s arguably selfish not to. Sharing data is only way the field moves forward.

What data actually does

The talk that stayed with me most: the AI-COLLAR study, led by Professor Valérie Chetboul at ENVA. 703 dogs, each wearing a biometric device invoxia, monitored continuously over several months in their real home environments. The goal was to establish proper baseline reference values for resting heart and respiratory rates in dogs, data that, didn’t exist before at this scale.
The results were striking enough that they’ve already changed clinical practice. Patterns became visible before any cardiac episode occurred. No human could have processed that volume of longitudinal data manually.

The Vet IN Tech Thesis Prize this year went to Alice Ruellan, a young veterinarian from ENVA, for work that deserves more attention than a prize certificate. Her pilot study used the same Invoxia Biotracker on a very different population: the working dogs of the GIGN – France’s elite intervention unit. These are Malinois operating in bite work, explosive detection, and high-stress environments. Their health directly affects mission success and operator safety.

A project followed a young working dog from an elite unit, wearing the same kind of device from his first day of training. When he began showing signs of overtraining months later – reduced performance, less expressivity, less appetite, the device data showed his nocturnal heart rate had risen by 10 bpm compared to his baseline. The signal was there the whole time, collected passively while he worked. The thesis frames it carefully as a pilot case, but the direction it points in is clear.

If you know what a dog looks like when healthy, you have a baseline for detecting when something is shifting, before it becomes a problem you can see. That’s not a distant possibility. It’s already happening, in the hands of researchers who are asking the right questions.

Tools that shouldn’t be reserved for the well-resourced

A veterinary ophthalmologist expert, whom I had a chance to know for a few years now, Bertrand Michaud, presented work comparing retinal imaging using a smartphone against expensive dedicated diagnostic equipment. The practical implication goes well beyond cost: many patients live hours from a specialist.
If a vet in a rural practice can capture a retinal image on a smartphone and share it with a colleague for review, the question of whether something is an emergency and whether that patient needs to make a four-hour drive tonight – becomes answerable.

There’s already a tool moving in this direction DogAEye, developed at the University of Pennsylvania, applying AI to retinal images in dogs. Veterinary telemedicine isn’t a concept anymore. It’s being built.

On AI, ethics, and the skills we can’t afford to lose

Two things the room seemed clear-eyed about. First: the AI revolution is happening regardless. Pretending otherwise doesn’t make it slower – it just means you have less say in how it unfolds. Better to be transparent about using it, to analyse the consequences openly, and to build proper structure around it – including in education.

In the spirit of full transparency: yes, AI helped me write this. All the reflections and thoughts are mine – it just made sure your eyes won’t bleed reading them.

Second: there’s already research on skill atrophy when AI handles tasks that humans used to do. A surgeon or vet who has always had a robotic assist may not be able to perform without one when something unexpected happens. The next generation of clinicians needs to learn the underlying skills, not just the assisted version of them. And AI, however capable, cannot yet handle the truly uncertain situations, the ethical edge cases, the moments that require empathy and judgment. That’s where the human remains not just present but necessary.

One more thing that was raised, and that I haven’t been able to stop thinking about. We are getting closer to tools that will let us understand the emotional states of animals: not just behaviour, but what’s actually happening underneath it. When that becomes possible at scale, we will be obliged to act on what we learn. Particularly for farm animals. That’s a conversation the field is going to have whether it’s ready for it or not.

There’s also a European sovereignty dimension that I keep hearing at tech conferences lately, and it came up here too. Most of the models and infrastructure we depend on are American. What happens if access changes? The question of building European alternatives – in France, in Poland (where I’m from, and where projects like Bielik are doing exactly this) – isn’t abstract anymore.

Finding my place in this room

I’ll be honest: I was quieter than I should have been. My French isn’t where I want it to be for this kind of environment, and I was still working out where I fit in a room of people whose professional world is quite different from mine, even if the questions we’re asking are often the same.

Next time I’ll talk more. But I left with a clearer sense that this is my crowd. Scientists who care about animals, who think critically, who are genuinely curious about what technology can do – without losing sight of what it can’t.

That’s worth coming back for.

woofLogs is building the behavioural data layer that doesn’t yet exist in canine health, what happens in training, over time, in the real world. If you’re working in animal health and want to talk: hello@wooflogs.com