early project, revisited · game AI · 2020 / 2026
Snake AI
Small Snake AI experiments, from older deep Q-learning code to newer checkpoint and evaluation work.
Snake is a useful toy problem because it is small enough to understand and just annoying enough to expose whether an agent is actually learning. I have multiple Snake folders: an older deep Q-learning project and a newer 2026 version with CNN checkpoints, evaluation runs, and model-visualisation tooling.
This is included as a learning trail, not because Snake itself is impressive. It shows an early interest in reinforcement learning, state representation, and reward shaping, then a later return to the same toy problem with better tooling.