Story 03 · AI Architect
Govern prompts, models, and datasets as first-class assets
Record the exact governed inputs behind an AI result so teams can trace identity, version, provenance, and relationships over time.
The situation
An architecture team needs to answer a basic but consequential question: which prompt, model, and dataset produced this result—and where did each one come from?
The question
Can we prove the exact versioned inputs and their provenance?
The governance workflow
From question to governed outcome
- 01
Register or observe the asset
Record a managed asset or submit immutable runtime observations from the system that owns authoring and serving.
- 02
Capture identity and provenance
Store version, source system, source reference, content hash, and other durable identifiers without requiring sensitive content.
- 03
Inspect the catalog
Use Studio to find the governed prompt, model, dataset, or evaluator-provider record.
- 04
Follow the relationships
Open ontology lineage to connect assets with evaluations, executions, decisions, and related governance records.
- 05
Use the asset as evidence
Reviewers can now determine the governed inputs behind a result without reconstructing history from separate systems.
In Studio
The screens behind the story
Prompt records establish a governed identity and version.
Model provenance is visible alongside its governed identity.
The ontology makes relationships between governed records explorable.
The outcome
The team gains a stable source of governance truth for AI inputs while each runtime keeps ownership of authoring, serving, and artifact storage.
OSS capabilities used