The Milestone Developer Summit 2026 returned to the historic Carlsberg Museum in Copenhagen for the second consecutive year. The summit brought together innovators, developers and technology partners in video technology. It also marked the second edition of the Hafnia Hackathon, where developers from around the world turned AI technologies into practical solutions to real-world challenges.
Edward Mauser, VP of Data and AI Services, said:
"With our new Data and AI Services we are giving developers the easy button for video analytics and vision language models"
Keynotes came from companies including NVIDIA, AWS, Dell and Intel. Alongside breakout workshops and the Project Hafnia Hackathon, they gave participants from around the globe a chance to share insights and predictions on what comes next.
THE CHALLENGE
Millions of video cameras capture footage that could improve everyday life. Much of that value goes unused, because reviewing countless hours of video takes too much time.
Participants were asked to find a real operational problem and turn video data into action. They built with Hafnia VLMs, NVIDIA Cosmos-3, the XProtect APIs and development tools. Their projects included intelligent alerts, natural language search, smart dashboards, camera health monitoring and entirely new applications.
FIRST ACCESS TO THE FUTURE
Before the summit, developers from 25 countries had access to the Hafnia VLM. It is built on NVIDIA Cosmos-3 and fine-tuned on real-world video from Hafnia's domain-specific data library, so it better understands city-specific visuals, languages, symbols, events, weather and lighting.
AND THE WINNER IS
Of the 16 teams, eight were shortlisted. A panel of judges from Milestone and NVIDIA selected the top three, and the audience voted for their favourite, making four winners in total.
Gabriel Crosnier de Bellaistre of Citilog won first place for Serve. Serve flags falls without anyone having to watch the video. A second AI questions the Hafnia VLM for more detail before an alert is raised.
Gabriel Crosnier de Bellaistre of Citilog won first place for Serve. Serve flags falls without anyone having to watch the video. A second AI questions the Hafnia VLM for more detail before an alert is raised.
Gabriel said:
“What I really wanted is to take from an experiment to a real-world application. So I built Serve to really help people, and the catchphrase we had is to serve and protect, and I really hope it will serve right and protect anyone needing it. I’m so happy. I lost some sleep, but it was worth it.”
He took home €5,000 and an NVIDIA Jetson Thor Developer Kit.
Roland Harwood, Community and Partnerships Lead, Milestone Data and AI Services, said:
"We were blown away by the innovations on display this year, where we more than doubled the engagement and impact of the hackathon. Huge congratulations to everyone who participated for their creativity and contributions and we look forward to continuing our collaboration with the community to shape the future of computer vision and responsible AI."
THE EIGHT FINALISTS
Serve. Gabriel Crosnier de Bellaistre, Citilog (1st Prize) - Flags falls without anyone having to watch the video. A second AI questions Hafnia before it escalates. From France. Watch Serve Demo Here.
Scenario Lab. Hugo Gonzalez Romo (2nd Prize) - A fire drill for a Hafnia alarm. It tests the alarm on generated video of the customer's own camera, before go-live. From Spain. Watch Scenario Lab Demo here.
Clearway. Hojin Park, Hanwha (3rd Prize) - Spots what is blocking streets and tracks, asks Hafnia once per incident, and sends it to the team that owns it. From South Korea. Watch Clearway Demo Here.
Trace Dump AI. Aymen Bennabi, Sneha Chaudhary, Mohsin Imam, Runtime Rebels, (Audience Favourite) - Handles illegal dumping from the first sighting to a verified cleanup. From Spain. Watch Trace Dump AI Demo Here.
VisualEYE. Elias Dargham, Ramy Abou Haydar, Systeminence (Finalist) - Puts a Hafnia check in front of a site's existing XProtect alarms, with an undo button, and can have Hafnia describe events. From Lebanon. Watch VisualEYE Demo Here.
Hafnia Dispatcher. Eric Bourque, SpecterDyne (Finalist) - Checks alarms with HAfnia, with faces masked before any clip leaves the site. From Switzerland. Watch Hafnia Dispatcher Demo Here.
Talos. Magnus Guldberg Pedersen, August Brogaard Tollerup, MAUG (Finalist) - The operator draws an area and names an object. Talos raises an XProtect alarm with the object's box. From Denmark. Watch Talos Demo Here.
Tollgate. Augusto Ferrarini, Ferrarini informatica (Finalist) - Decides, clip by clip, whether a Hafnia request is worth its cost. From Italy and Brazil. Watch Tollgate Demo Here.
