PLARENA
Students don't watch an algorithm — they run it, break it, race it, and write their own agents against it. Every arena is one self-contained HTML file: no accounts, no servers, no setup. A share-link with a seed is the assignment.
Search Arena live
Mazes, cities, and coverage worlds. BFS, DFS, Greedy, Dijkstra, A*, and hill climbing with a probe-level search tree, a live frontier data-structure panel, fog of war, custom heuristics, and a student-code agent API.
Game Arena live · v0
Minimax and alpha-beta pruning on tic-tac-toe: the game tree drawn live with mini-board nodes, pruned branches shown, a real-time α/β window, move-ordering experiments, a student-code agent API, Kriegspiel (blind) mode, and stochastic opponents. Play the optimal opponent — or be the flaw in its assumption.
ML Arena live · v0
A tiny recommender where your clicks are the training set: rate Miami dishes, then watch linear regression, logistic regression, and an MLP learn your taste on one canvas. Break them with an XOR palate, cause overfitting on purpose, bet on the winner before training reveals it. A second entry mode re-skins the same three models onto a student-grades dataset.
Algo Arena live · v0
One 12×12 map, nine rooms, one story: escapes from brute force. BFS rings to A*'s certified failures, backtracking on a stubborn 5-cycle, greedy with a real exchange argument, the map turning into a DP table, max-flow at a visible bottleneck — and a finale where 720 tours are easy and 2.4×10¹⁸ are not. A growing Scoreboard of exact, replayable counts is the course summary.
Vision Arena planned
Convolutions, edges, and features on images students bring themselves.
Data Arena planned
Clustering and frequent patterns as a playground, not a slide.