Cookbook — From Customer Problem to CAIP Recipe
Connected AI Platform (CAIP) has grown into a broad set of offerings across three evolution stages (Managed Kubeflow and CAIP 2.0), and it isn't always obvious which one solves your problem. This section flips the perspective: instead of organizing by product, it organizes by customer challenge, served up as a set of ready-to-follow recipes.
Each recipe below states a problem a team typically comes to us with, lists the ingredients — the CAIP capabilities you'll need — and then walks through the steps to combine them, with links straight into the relevant product documentation.
Find the row below that best matches what you're trying to do, then open that recipe. If your challenge spans multiple rows (e.g. you need both training and inference), open both — the recipes cross-link where ingredients overlap.
Find your recipe
| Customer challenge | Recipe |
|---|---|
| "We need to extract, transform, and load large volumes of data, and orchestrate that processing as repeatable workflows." | Data Processing & Orchestration |
| "We need to train and track classic ML models, or train / fine-tune LLMs at scale, and keep our models in a registry." | Model Training & Model Registry |
| "We need to serve our models for predictions — in real time or in batch — including our own self-hosted LLMs." | Model & LLM Inference |
| "We want to build AI agents, chatbots, or RAG-based apps on top of LLMs, and observe how they behave in production." | Agentic AI & LLM Applications |
| "We're running on the retired AI-Research cluster / classic Kubeflow pipelines and need to move onto a supported CAIP offering." | Migration Support |
Why this cookbook exists
Customers frequently know their business problem well but aren't sure which part of CAIP — Managed Kubeflow, CAIP 2.0 Workflows, the LLM API, Agents Foundry, Inference, Apps, and so on — actually addresses it. These recipes exist to close that gap: they are the fastest way to go from "I have a problem" to "here's the CAIP feature, and here's how to start using it."
If your challenge isn't covered yet, reach out via Support — we continuously expand this cookbook as new customer problems and platform capabilities emerge.