watsonx Platform Fundamentals and Architecture
This domain tests whether you can place each watsonx component in the architecture and explain how work is scoped, secured, and moved between them.
What each product does
- watsonx.ai - authoring and runtime. Projects hold notebooks, Prompt Lab work, experiments and models. Deployment spaces hold deployable assets and their endpoints.
- watsonx.data - the open lakehouse: Presto/Trino for federated SQL, Spark for ETL and ML, Apache Iceberg as table format, a Hive metastore, and object storage such as IBM Cloud Object Storage or S3.
- watsonx.governance - AI Factsheets, model inventory, risk assessment, evaluation and monitoring.
- watsonx Orchestrate - adjacent product for agents and automation skills.
Scoping and identity
The hierarchy runs IBM Cloud account → resource group → service instance → project or deployment space, with catalogs sharing assets across projects. IBM Cloud IAM roles and Cloud Pak for Data platform roles decide who can create, edit and promote assets. The platform runtime is Red Hat OpenShift, available as a managed IBM Cloud service or self-managed.
Worked example
A team trains a churn model and must prove what data it saw. Reasoning: the notebook runs in a watsonx.ai project, so experiments, training data references and metadata are tracked; the model is promoted to a deployment space for serving; registering it in the governance inventory produces a factsheet recording training data, metrics and lineage. An answer that stops at "deploy" misses the governance half of the requirement.
Traps
Assuming watsonx.data stores the bytes (it queries data that stays in object storage); confusing a project with a deployment space; treating governance as post-deployment paperwork rather than a lifecycle record; assuming every service shares one IAM role model.