In the first part of “How to Teach Your Own AI Model,” we examined what happens beneath the surface of model training, from datasets and tokenization to alignment and inference. Now, Piotr Miłkowski returns to continue the conversation with host Kewin Wereszczyński, turning those foundations into practical guidance for developers.

Part 2 focuses on what you can realistically build and train today. Do you need millions of dollars? Is one GPU enough to learn something useful? Piotr and Kewin will discuss common beginner mistakes, the limits of training as an individual developer, and how to decide whether a project needs a custom model, fine-tuning, or an existing API.

The stream will also explore the changing role of training data. You will learn where synthetic data can help, why human-created data still matters, and whether the industry is approaching a shortage of useful material for training increasingly capable models.

Join us on Thursday, July 16, 2026, for a practical follow-up aimed at developers ready to move from understanding AI training to trying it themselves. You will leave with project ideas, clearer technical trade-offs, and one concrete experiment you can start with a weekend and a GPU.