Agents That Learn: A Self-Improving AI Hackathon
Most AI agents are only as good as the day they shipped. This hackathon is about agents that get better on their own.
Join Charlotte's builders for a one-day hackathon on self-improving AI agents: agents that learn from feedback, evaluate their own outputs, refine their prompts and tools, and remember what worked. Form a team, build against the clock, and demo your agent to a panel of judges for a chance to win prizes.
The challenge
Build an agent that measurably improves over time. Some directions to explore:
Feedback loops that turn user corrections into better behavior
Self-evaluation and reflection, so the agent critiques and retries its own work
Automatic prompt and tool optimization driven by evals
Long-term memory that helps the agent learn from past runs
Observability and traces that prove the improvement
Rules
Teams of 1–4. Solo builders are welcome, and we'll help you find teammates.
All project code must be written during the event. Open-source libraries, frameworks and APIs are fine.
Submit a repo link and a short description by 3:30 PM.
Demos are 3 minutes per team.
Judges' decisions are final.
Who should come
Developers, ML engineers, data scientists, students and curious builders at any level.
Bring
A laptop and charger, and your API keys if you have them. We'll provide lunch, snacks, Wi-Fi, and prizes.
Schedule
10:00 AM: Check-in, coffee and team formation
10:30 AM: Kickoff, challenge brief and rules
10:45 AM: Hacking begins
12:30 PM: Lunch (working lunch, mentors on the floor)
2:00 PM: Mentor check-in round
3:30 PM: Code freeze and submissions
3:45 PM: Team demos
4:30 PM: Judges deliberate, networking
4:45 PM: Awards and group photo
5:00 PM: Event ends