How do you get high-school students to build an AI that solves a maze it cannot even see?

That was our challenge at the Bocconi Summer School, where our representatives Dan Anghel, Roberta Claps, and Gaukhar Abdiganiyeva ran two sessions this week.

First the theory — nodes, edges, BFS, Dijkstra, A*. Then the real thing: build an agent that reaches the goal in a 50x50 maze with invisible walls. It only ever sees its four neighbouring squares, so it has to map the maze as it moves and re-plan every single turn. Six rounds, teams, a live Kaggle leaderboard.

Watching them go from their first graph to debugging a pathfinding agent was the best part of the week.

Thank you to Università Bocconi for hosting us, and a special thank you to Omiros Papaspiliopoulos, whose trust and support made this possible.

Connecting minds, not only layers.