This weekend, I had the opportunity to mentor participants and serve as a judge for the Champions' Choice track at NVIDIA Spark Hack Toronto.
The hackathon brought together engineers, researchers, founders, and students to build agentic AI applications using open models and Toronto open data. Teams explored challenges in public services, urban operations, and economic systems, with projects running on NVIDIA DGX Spark systems powered by the GB10 Grace Blackwell Superchip.
Over the course of the weekend, I reviewed projects that ranged from helping residents navigate city services to improving decision making around transportation, infrastructure, and local economic activity.
One thing I noticed across many of the strongest submissions was the amount of thought that went into the overall system. Teams spent time deciding what tasks should be automated, what information should be retrieved, when agents should call tools, and when they should ask for human input.
The projects that stood out were usually easy to understand. They solved a specific problem, had a clear user in mind, and demonstrated how the system would work in practice. The teams could explain why they chose a particular architecture and where the limitations were.
It was also encouraging to see how quickly builders are adopting agentic workflows. Just a year ago, many hackathon projects centered on chat interfaces. This weekend, teams were building systems that could reason across multiple steps, interact with tools, and take action based on real-world data.
Thank you to NVIDIA, the organizers, mentors, judges, and participants for a great event. I enjoyed meeting so many talented builders and look forward to seeing where these projects go next.
Top comments (0)