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

Most AI engineers are not training frontier models from scratch. That takes a large research team, substantial infrastructure, and a large budget. The more practical goal is to understand what each stage contributes, what it cannot fix, and how those choices show up in production behavior.

This chapter walks through the pipeline from raw data to a model that is ready to evaluate and deploy.