In this talk, we will examine how to decompose AI systems into more manageable parts that then can be independently developed and tested, and then easily be composed together into an AI system. We will present a unified architecture for building batch, real-time, and LLM AI systems around 3 classes of machine learning pipelines: feature pipelines, training pipelines, and inference pipelines.

Just like you can make great music with 3 chords, we will show tens of examples of great AI systems built with our 3 ML pipelines (and the truth!).

We will show how our 3-pipeline architecture helps align teams and accelerates time to value and quality.