I know it's become a bit of a tradition to open these announcements by saying how impossibly hard it was to pick winners, but this time it's not even slightly exaggerated. Smarter Spaces asked people to embed intelligence into the places they live, work, and play, and what came back was genuinely some of the most creative, ambitious, and well-executed edge AI work we've seen from the community.
We put the projects in front of a panel of judges from across both Axelera AI and our good buddies at Orange Pi, and opened company-wide voting to help narrow things down. Even then, the top places were all tied and we had to press gang some more judges into action to help determine the winners - it really was that close! If your name isn't in the top three, know that you were right there in a heated conversation.
Huge thanks to our friends at Orange Pi for partnering with us on Smarter Spaces. The Orange Pi 5 Plus boards were a core part of every participant's kit, and the combo of Orange Pi + Metis proved to be a seriously capable edge AI platform across all eight projects.
Before we get into the full roundup, please join me in congratulating the winners...
🏆 1st Place: EUR 1,000
Project GAIUS, by @David
Iron Man-style smart home control: point at a device, make a gesture, control it.
Excelsior! David built a full 3D spatial engine with multi-camera triangulation, arm-based raycasting, a visual room designer, and direct LG TV integration. One community member commented: "I thought we are working on a hackathon not a full fledged ready-to-market product." Well done on a technically brilliant, but also brilliantly creative and unique approach to HMI control! ‘Nuff said.
🏃🏽Runner Up: EUR 500
MotionFlow, by @FreezerJohn
A privacy-first activity sensor that classifies what people are doing from pose geometry, with zero video leaving the device.
Jonathan's project runs YOLOv11-Pose on Metis, tracks people, works out what they're doing (sitting, walking, entering a room), and publishes events over MQTT. No video, no cloud, no compromise. The clincher? It's been running 24/7 in his actual home for days. Judges highlighted the cross-industry relevance, from elder care to occupational health. "Use case is relevant across a number of industries," noted one judge. Real-world validation like this is hard to argue with.
🤖 Runner Up: EUR 500
NEMA MedBot, by @Vito
An on-device identity recognition and posture analysis pipeline targeting clinical healthcare environments.
Vito teaches robotics, electronics, and IT at a technical institute in Brescia, Italy, and has a commercial partner already identified for clinical validation. Judges highlighted the commercial readiness and powerful market fit. This isn't just a challenge project. It's an R&D initiative with a real-world deployment roadmap and students involved in the build.
🌟 Honourable Mentions and Community Highlights
I’ll get in touch with the winners about the prizes 👍 But every one of the Smarter Spaces projects deserves attention. These aren't quick weekend hacks. These are multi-week engineering efforts with real hardware, real inference, real code, and in many cases, real deployments running in people's actual homes and workplaces.
Remember: the people judging these projects build AI systems, pipelines, and products professionally, every single day. When they say they're impressed, it really means something!
📚 A.B.C. (AI Book Cataloguer), by @DavideS
A tabletop station that scans book covers and automatically extracts title, author, and publisher using on-device OCR and object recognition.
The ITLAMP team in Italy built something with immediately obvious commercial applications. Libraries, archives, second-hand bookshops, warehouse inventory. The judges praised the creative use of OCR on Metis and the highly polished implementation. This was also one of the most prolific projects in terms of community engagement, with four separate build update posts documenting the full journey from concept to working prototype, which is real dedication.
🚨 FlowSentry-Wake, by @Enmin and @mm0806son
A security system that fuses optical flow with YOLOv8 to catch intruders even when physically camouflaged.
Two KU Leuven students solved three separate compiler-level challenges to become what we believe are the first people to get optical flow running on Metis. They tested it by literally crawling under a blanket to evade standard object detection, and the system caught them anyway! That's not a demo, that's a product feature. There were early questions as to whether this might "stretch the compute", but the guys proved it works.
🏥 RehabVision, by @dilip.m
A physiotherapy assistant that uses Metis-powered pose estimation to guide patients through rehab exercises at home.
RehabVision counts reps, checks form, and gives real-time feedback, all locally, with no health data leaving the device. Judges flagged the "huge market" potential, and they're right. Millions of people do physio between clinic visits with little to no feedback on whether they're doing the exercises correctly. This addresses a real gap, need and a problem that’s now been solved.
🔐
SafeHaven, by @imprettychill_
A privacy-first home security system that answers the questions most NVRs don't: is the door locked? The garage closed? The gate secured?
SafeHaven converts Metis-powered visual inference into semantic security events and pushes them into Frigate NVR and Home Assistant, all offline with a full evidence timeline. Shashi was a first-time community participant, and the integration with Frigate and Home Assistant was praised by judges as exactly the kind of ecosystem thinking that demonstrated how the future is edge AI.
🖥️ Smarter Desk, by @carlo
A gesture-controlled desk assistant that uses hand detection on Metis to trigger screenshots for feeding future AI agents.
Carlo managed to build this while literally touring Germany with his entire kit in a bag, which honestly deserves a special award for portable edge AI development. The project showcases what even a focused, compact use of Metis can achieve when you apply it to your daily workflow.
💬 Final Thoughts
We’ve seen some completely different interpretations of what "smarter spaces" means. From Iron Man-style gesture control to physiotherapy assistants, from optical-flow intruder detection to automated book cataloguing. The range is extraordinary, and every single project was built, documented, and shared openly with this community.
A few things stood out across the board. Privacy-first, offline-first design wasn't just a checkbox. It was a core philosophy. And the build updates posted throughout the challenge created some of the most valuable technical content this community has ever produced.
To everyone who submitted a project, posted updates, asked questions, or left encouraging comments on other people's work: you've shown exactly what this community is about.
We've got more challenges, campaigns, and activities coming up, so keep building, keep sharing, and keep making your spaces smarter!
