The model you use through an API is not the result of one training job. It is the result of a pipeline: data collection, filtering, pretraining, instruction tuning, preference optimization, safety work, evaluation, and deployment-specific adaptation.

For most AI engineers, the goal is not to train a frontier model from scratch. That is a large research and infrastructure program. The goal is to understand what each training stage contributes, what it cannot fix, and how those choices show up in production behavior.

This chapter walks through each stage of that pipeline.