learning in public

Building an LLM
from Scratch

Zero CS background. RTX 5060. Every lesson written, committed, and shipped publicly — from next-word prediction all the way to training a working transformer. No shortcuts. No skipping the math.

Start Lesson 1 → ⭐ Star on GitHub

1 days

4 lessons done

0 CS degree

Phase 1 — Foundations

01 What Is an LLM? done

next-word prediction · parameters · product mental model

02 Python from Zero done

variables · lists · functions · loops · tiny fake LLM

03 Vectors & Matrices done

embeddings · king−man+woman≈queen · tensor shapes

04 PyTorch from Zero done

tensors in code · autograd · first training loop · nn.Linear

05 Build a Tiny GPT next

attention · transformer blocks · train on GPU

Phase 2 — Real LLMs

06 Run LLMs Locally with Ollama soon

Llama 3.1 on RTX 5060 · quantisation · prompt via terminal

07 Build Your First LLM App soon

Python · API · input → LLM → output

Building an LLMfrom Scratch

Building an LLM
from Scratch