Everything Is Matrices
How Language Models See, Think, and Act
By Sree Charan Reddy
Start Reading →The Basics
Vectors, dot products, neural networks, backpropagation. Everything built from first principles with real numbers.
The Transformer
Attention, multi-head attention, feed-forward networks, layer norm. Every component traced step by step.
Scaling & Training
MoE architectures, scaling laws, pre-training on trillions of tokens, RLHF, and extended thinking.
Inference
Token-by-token generation, KV cache, prompt caching, and how models handle million-token contexts.
Multimodal & Agents
Vision encoders, native multimodal models, tool use, MCP protocol, and multi-step agents.
Hands-On
Build a transformer from scratch, fine-tune an open model, and build a working agent with tool use.
Made with ♥ by Sree Charan Reddy