It is finally the last installment! By the end of the last part, the functionality was complete. However, as it stands, it requires typing commands in the terminal, which is a bit of a high barrier to ...

A reinforcement learning framework for automatic game control. The agent learns to play games by observing screen pixels and outputting keyboard/mouse actions, using a Dueling DQN architecture with ...

Real-time English subtitle translator for foreign-language football broadcasts. Watches audio from any football stream, auto-detects the language, and displays rolling English subtitles — with no ...

Let local LLMs condense and structure messy ideas before sending to Claude to save tokens and time. Keep a manual copy-and-paste checkpoint to refine specs; it prevents overcomplicating and wasted ...

I had completed everything up to the Transformer block, which combines tokenization, Embedding, Self-Attention, LayerNorm, residual connections, and FeedForward. The core mechanisms of GPT were all in ...