There was something quietly electrifying about watching the Carlsberg Museum slip from its usual calm into a pulsing hub of video technologists, AI researchers, platform architects, and a whole crowd of developers who looked like they were genuinely hungry for the future. The place has an odd gravity to it, partly because of its scientific heritage — this is where the pH scale was born, after all — and partly because it carries a kind of “ideas have been made here” energy that fit strangely well with the Hackathon keyboards clacking under frescoed ceilings.
The summit itself unfolded like a declaration that video technology is now sprinting into its next era, and Milestone clearly intends to be the one holding the door open for everyone else. When Sebastian Döllner took the stage and spoke about open platforms not just as business strategy but as an ecosystem of shared invention, the message resonated more like a manifesto than a keynote. And frankly, you could feel it in the room — nobody believes a single vendor can solve the puzzle of smart cities, responsible AI, and global-scale video data alone. The ones who thrive will be the ones who collaborate aggressively, borrow brilliance, and let others build on top of their foundations.
Not surprisingly, the lineup of speakers matched that ambition. NVIDIA, AWS, Dell, Intel… all contributing different angles of where compute, cloud, content creation and AI reasoning are headed. Breakout rooms buzzed with everything from responsible AI discussions to the almost philosophical debates around how much autonomy we really want to give video analytics in the public domain. And woven through all of it was the Hafnia Hackathon, which quickly became its own little planet of rapid prototyping, whiteboard chaos, and API experimentation.
The real fireworks, though, happened before anyone even stepped into the hall. AI developers from fifteen countries got early access to Milestone’s upcoming specialized Vision-Language Model, built on NVIDIA’s Cosmos-Reason and then painstakingly post-trained on Hafnia’s enormous annotated city-specific dataset. The result? A model that doesn’t just “see” — it understands the texture of Copenhagen fog, the semantics of an icy Nordic street sign, the rhythm of a bike lane at rush hour. That gave the Hackathon challenge real teeth: build integrations on top of this new VLM via API and try to imagine what a truly smart city system could look like when machine perception finally catches up to messy human reality.
Milestone, meanwhile, used that momentum to preview its own generative AI-powered plugin for XProtect — one of those quietly revolutionary features that will probably become standard in a few years. The idea is simple: instead of drowning operators with raw footage, the plugin turns video into structured written reports, synthesizes activity, and produces validated real-time alerts. Traffic management for cities, ports, airports — basically every environment clogged with sensors — suddenly looks a lot more manageable, and a lot more scalable.
But the best part of events like this is always the people who show up and build things that no one quite expected. Of the six shortlisted finalists, the winner — Thomas Kreutz — created something that genuinely caught people’s imagination. His “Ask The City” project feels like a little shard of the future: pick a spot on the map, ask a question in natural language, and the system answers based only on what it sees in the last few seconds of live video. No identity leakage, no overreach, just a crisp, privacy-aware interpretation of real-time urban life. Hearing him reflect on how easy it was to get started with the API made it clear that this wasn’t a staged showpiece — he built something real, fast, because the tools finally allow for it.
The rest of the finalists filled in a kind of panorama of what’s possible. Emergency response triage powered by edge AI. Unified incident maps that blend weather, traffic, and visibility into actionable overlays. No-code workflow builders for integrators who want AI analytics without the engineering overhead. Semantic clip summarizers. Multimedia intelligence platforms that make every frame queryable. All different, yet all anchored in this new idea of video that isn’t passive anymore — it’s becoming an active intelligence layer stitched into the physical world.
Walking out of the Carlsberg Museum at the end, with the evening settling over Copenhagen, there was this lingering sense that the summit wasn’t just another developer event. It felt more like a marker — the point where open platforms, specialized VLMs, and practical AI moved from experiments into something operational, deployable, and imminently real. And if the energy of this year’s hackathon is any indication, next year’s innovations won’t just be incremental. They’ll be wild, unexpected, maybe even a little disruptive — the good kind of disruptive that pushes an entire industry forward.
Leave a Reply