Last weekend we were delighted to sponsor Nova, Carnegie Mellon's premier GenAI hackathon, hosted by the awesome team at ScottyLabs. The mission for the 248 students packed into CMU's Swartz Center was simple but ambitious: build something innovative with AI in just seven hours.
We showed up with $500 in cash, $500 in Agentuity credits, and a booth full of stickers. 31 teams, out of a total of 53, signed up for our 'Best Use of Agentic AI' track. As both sponsors and judges for our track, we spent the day engaging with students — answering questions, discussing ideas, and evaluating projects that showcased genuine curiosity about building AI agents in production.
What a Seven-Hour Sprint Can Build: The Best Agentuity Teams
We were blown away by what teams were able to build with Agentuity under such tight time constraints. Seven teams built and deployed functional projects, giving us a glimpse of just how much you can do with our platform:
- CactusTech: Built a sophisticated memory management system for agents, with a team of Cognitive Science majors blending their understanding of the human brain with code to explore short, medium, and long-term agent memory. Their solution showcases a deep dive into key features of our platform, including vector storage, KV storage, and logging.
- Nova AI Memory System: In a fantastic meta approach, this team built an agentic CRM for the event's host, ScottyLabs. The system is designed to learn how the organization matches students with projects, effectively capturing institutional knowledge — see their implementation.
- DAM: Created a debate platform where agents and humans can argue the nuances of research papers, providing a fascinating look at agent-to-agent communication.
- Nova Agent Project: Transformed structured learning documents into engaging, interactive quizzes. See how they built it.
- LaunchKit: An agent that generates essential marketing materials to help new companies launch. Explore their code.
- Recipe Generator: A clever tool that creates recipes based on a photo of the ingredients you have in your fridge. See how they implemented it.
- SNAP Map: A mapping application for visualizing SNAP benefits and accessibility. See the technical details.
- ImpactUs: An app built to help event organizers understand their impact from the attendees' perspective.
And the Winner is… Language Learning, Reimagined
For our "Best Use of Agentic AI" track, we chose Immersion — a language learning platform that uses AI-powered learning experiences to help users master new languages. What impressed us was their technical execution and thoughtful approach to building agentic AI. The team built a sophisticated multimodal agent that combines real-time voice interaction and animated avatars. A custom "ConversationCoach" intelligently detects when learners hesitate, provides native-language nudges, then guides them back to the target language.
Their project featured a 10-stage pipeline complete with role-play scenarios and visual stimuli like restaurant menus. It was one of the most creative agent use cases we saw. Explore their code.
Their work exemplifies exactly the kind of agentic AI we're building infrastructure for. With this choice, we're making a bet on talent: bring exceptional developers into the Agentuity ecosystem, give them resources to build, and learn from what they create. We're excited to support their work and learn from their feedback as they continue building.
Real-World Use Cases and Building Connections
Throughout the day, we kept hearing the same question: "Can I process real-time video with this?" Multiple teams wanted to build agents that could analyze live video feeds — everything from accessibility tools to real-time translation. We also saw teams tackling memory management (how does an agent remember what matters?), multi-agent coordination, and personalized workflows through OAuth integrations.
When we explained Agentuity to judges and industry professionals, one analogy clicked instantly: "OpenAI Builder is like Wix or Squarespace, n8n is like WordPress, and Agentuity is like React or Next.js for agents." They immediately understood what we're building and why it matters.
These conversations, with both students and professionals alike, are invaluable to us. The widespread interest in real-time video, for example, reinforces our commitment to powerful features like our comprehensive streaming support, and directly informs how we can make these capabilities even more accessible.
But the value of this event extends beyond product feedback. It was about building genuine connections within the Carnegie Mellon developer community. CMU is one of the world's top Computer Science schools and has students majoring specifically in AI Engineering, which is our ideal target demographic. Professors and judges showed strong interest in Agentuity's approach, and we're excited about the potential for long-term collaboration in this space.
What's Next
The feedback and energy from Nova is already shaping our roadmap. We're more committed than ever to refining the Agentuity experience for fast-paced, high-pressure environments like hackathons. We'll continue supporting the students and teams that build with Agentuity, showcasing their work as they develop new and exciting projects. We're also excited about deepening our collaboration with ScottyLabs and the broader CMU developer community. In other words — Nova 2025 was just the beginning!
Acknowledgments
Thank you to ScottyLabs for hosting an incredible event and welcoming Agentuity into the CMU community. Thank you to Carnegie Mellon University, the judges, and all the students who brought amazing creativity and technical skill to Nova.
Finally, to CactusTech, Nova AI Memory System, DAM, Nova Agent Project, LaunchKit, Recipe Generator, SNAP Map, and ImpactUs — thank you for building with us. What you created was impressive, and we can't wait to see what you do next!