Back to projects

LLM education // transformer // model internals

Recreating ChatGPT from scratch

Educational build that reconstructs the core blocks of a generative chatbot: tokenization, training, inference and model limits.

The project

This project reconstructs the essential blocks of a conversational model to explain the architecture behind systems people casually call ChatGPT.

Technical angle

The build goes through text preparation, tokenization, embeddings, transformer blocks, training loop and inference.

What it proves

It shows the ability to explain a complex system by rebuilding enough of it for its constraints to become concrete.