Build a Reasoning Model (From Scratch)

You will learn to:

1. implement key improvements for reasoning in LLM;
2. evaluate models using expert judgments and benchmarks;
3. enhance the ability to reason without retraining weights;
4. connect external tools (e.g., calculator) through RL;
5. apply knowledge distillation from larger reasoning models;
6. understand and build a complete development pipeline for reasoning models.

Reasoning models break down tasks into steps and provide more reliable answers in mathematics, logic, and programming - an approach already used in leading systems like Grok 4 and GPT-5. This course demystifies the process: you will start with a small foundational LLM, incrementally add reasoning mechanisms, learn to measure real quality improvements, and then further enhance it using non-training methods and RL. By the end, you will have a compact but capable reasoning stack, created by your own hands.