Prompting Granite for “JSON / YAML Mode”

[aakriti1318]

Structured outputs are foundational for production systems. Teams adopting Granite need reliable, machine-parsable responses for APIs, agents, workflows, and enterprise automation.

Potential points to cover

• How to prompt Granite to reliably emit JSON/YAML only
• Define schemas in prompts
• Validate responses and handle errors
• Integrate with code (Python/JS)
• Using structured outputs to drive tool calls, database writes, and downstream services

[aakriti1318]

Assigned to @aakriti1318

[bjhargrave]

Isn't this mostly constrained decoding? e.g. with_structured_outputs in LangChain?

[bjhargrave]

Is this really agent specific? We already have recipes in this space in the snack cookbook. I don't think we want to duplicate work in this repo which should focus on agent oriented topics and not general gen AI topics.