You need to design a customized deep neural network in Keras that will predict customer purchases based on their purchase history. You want to explore model performance using multiple model architectures, store training data, and be able to compare the evaluation metrics in the same dashboard. What should you do?

The core requirement is to explore multiple custom model architectures and compare their evaluation metrics in a unified dashboard. Vertex AI Pipelines (the managed version of Kubeflow Pipelines on Google Cloud) is designed for this exact purpose. By creating an "Experiment," you can group multiple pipeline "Runs," where each run can train a different model architecture. The Vertex AI Experiments UI provides a built-in dashboard to directly compare the parameters and evaluation metrics (e.g., accuracy, loss) of these runs side-by-side, fulfilling all stated requirements.

A. AutoML Tables automatically builds models; it does not allow for designing a customized deep neural network in Keras, which is a specific requirement of the question.
B. Cloud Composer is a general-purpose workflow orchestrator. While it can trigger training jobs, it lacks a native, integrated dashboard for comparing ML-specific evaluation metrics across runs.
C. Running multiple jobs on Vertex AI Training is a valid step, but simply using similar names does not provide a comparative dashboard. You would need an additional tool like TensorBoard for comparison.

1. Google Cloud Documentation, "Introduction to Vertex AI Pipelines": "You can use Vertex AI Pipelines to orchestrate your machine learning (ML) workflows... By grouping runs in an experiment, you can compare the results of different pipeline configurations, such as different hyperparameter settings." This supports the use of pipelines and experiments for comparison.

Source: Google Cloud Documentation, "Introduction to Vertex AI Pipelines".

2. Google Cloud Documentation, "Introduction to Vertex AI Experiments": "Vertex AI Experiments is a tool that helps you track and analyze different model architectures, hyperparameters, and training data... You can compare experiment runs by selecting them in the Google Cloud console and viewing a comparison table of their parameters and metrics." This directly confirms that Experiments provide the required comparison dashboard.

Source: Google Cloud Documentation, "Introduction to Vertex AI Experiments".

3. Google Cloud Documentation, "Introduction to AutoML Tables": "AutoML Tables enables your entire team... to automatically build and deploy state-of-the-art machine learning models on structured data." This highlights that AutoML is an automated process, which contradicts the user's need to design a custom Keras network.

Source: Google Cloud Documentation, "Introduction to AutoML Tables".