The foundation of any successful AI project is high-quality, reliable data. The SAP AI lifecycle emphasizes that data preparation and pipeline establishment are prerequisites for model training and deployment. Starting with messy data will lead to poor model performance (garbage in, garbage out). Options 0, 1, and 3 skip this essential foundational step.

In SAP BTP, services like SAP AI Core operate with resource quotas defined in service plans. A ‘quota exceeded’ error typically relates to the BTP entitlements and resource group assignments, not the application code. The first step is to check the service instance binding and the associated quota plan in the BTP cockpit. While 1 and 4 are possible, they are less likely to produce this specific system-level error message.

In SAP AI Core terminology, a ‘Deployment’ is the process of creating a serving endpoint for a trained model. It takes a model URL and a serving configuration to launch a service that provides inference APIs. A ‘Configuration’ defines parameters for training or serving. An ‘Execution’ is a run instance of a workflow. A ‘Scenario’ is a logical grouping of AI use cases.

SAP AI Core’s ‘Schedules’ feature allows users to create automated, recurring executions of training or inference workflows. You can define a cron-like schedule to trigger the workflow periodically, which is ideal for continuous retraining scenarios. The other options are core components but do not handle automation scheduling.

The model registry (or model repository) in SAP AI Core provides governance and traceability. It stores model artifacts (like the .h5 or .pkl file) with metadata (name, version, metrics). This allows teams to reliably select, track, and deploy a specific version of a model, enabling MLOps best practices like version control and rollback.

SAP AI Core executes workflows within Docker containers. The standard method for including custom dependencies is to list them in a `requirements.txt` file placed in the same directory as your main training/inference script. This file is processed when building the Docker image for your workflow. AI Core does not support dynamic internet fetching or global server installations for user workflows.

This reflects the two primary user personas. ‘Workspaces’ in AI Launchpad provide a technical interface for AI practitioners (data scientists, ML engineers) to manage the full lifecycle—executions, deployments, models, and resources. ‘Applications’ are tailored, simplified interfaces built for specific business roles (e.g., a ‘Demand Forecasting’ app for a supply chain manager) that hide the underlying complexity.

An ‘Execution’ is the definitive record of a single run of a workflow. It captures a snapshot of all inputs: the workflow (code), the configuration (hyperparameters, dataset paths), the start/end time, status, logs, and the output (e.g., the model URL). This provides full traceability and reproducibility for each model training or batch inference job.

SAP AI Core provides a ‘Secrets’ feature specifically for managing sensitive data like passwords, API keys, and cloud storage credentials. You create a secret via the API or BTP cockpit, and then reference its name in your workflow configuration. This method avoids exposing secrets in code or configurations and leverages SAP BTP’s secure storage.

The foundation of any successful AI project is high-quality, reliable data. The SAP AI lifecycle emphasizes that data preparation and pipeline establishment are prerequisites for model training and deployment. Starting with messy data will lead to poor model performance (garbage in, garbage out). Options 0, 1, and 3 skip this essential foundational step.0