benwaar/games：用 PyTorch 从零构建游戏 AI 与机器学习项目集

Hands-on machine learning and game AI projects — reinforcement learning, deep learning, audio classification, neural networks, and model distillation. Built from scratch in Python with PyTorch.

Development Ethos

These projects are about:

understanding algorithms by building them
seeing how different approaches think
learning by stripping problems down to their nuts and bolts

Everything is text-based, inspectable, and hackable.

Projects

Utala: KAOS 9 — card game AI: TD learning, deep RL, distillation ✅
Signal Hunt — audio classifier: data pipelines, CNN, transfer learning, chords, CNN-RNN ✅
Incant — spec-driven code gen: local LLM + RAG → WAT/WASM + Z80 assembly 🔨
Acoustic Odyssey: cast — edge deployment: ONNX, quantisation, sub-100ms on-device inference
Acoustic Odyssey: echo — adaptive RL: personalises task difficulty to the user
Artefact: dig — binary art scanner: find hidden sprites in ROM using CV and autoencoders
Artefact: bloom — pixel art upscaler: super-resolution CNN and GAN
Artefact: dream — LoRA fine-tune a diffusion model on extracted sprites
Void Duel — ZX Spectrum shooter: classic ML, multi-agent RL, extreme-edge distillation

Utala: KAOS 9

A competitive 2-player tactical duel playable with any standard 52-card deck.

AI research studying skill expression, risk management, and learning algorithms — from random baselines through hand-built TD learning to deep reinforcement learning and model distillation.

Utala KAOS 9 Rules (PDF) · Project folder

Explore the Aerythen Universe

Utala: KAOS 9 contains artificial intelligence code for the card game in the Aerythen universe — a Command Line Punk world created by David Benoy.

The overarching project bridges art, music, AI and a developing series of companion novels and games.

🎮 Play the demo: Aerythen Interactive Demo
🎵 Listen to the Soundtrack: Aerythen Soundtrack
📖 Read the novel: Utala — An Aerythen Novel

Signal Hunt

A deep learning project — from raw audio to a trained classifier.

Turn raw audio (hums, whistles, claps) into clean Mel-spectrogram tensors, then train a hybrid CNN-RNN to classify them.

Project folder

License

Source code is licensed under the MIT License.