How to use AI assistance to draft a model in Yantra Studio
Use the assistant that matches your starting material to produce a first draft, then review it before any of it becomes the model.
When this is useful
- You have an OpenAPI specification or a body of SQL and would rather not transcribe it.
- A blank page is the obstacle and a draft to argue with would be more useful than a blank one.
- You want a second opinion on a design you have already drawn.
Before you start
- Whatever material you are starting from, and the understanding that the output is a proposal presented for review rather than a change applied to your model.
Pick the assistant that matches your input
The AI Model Assistant works on the data model. The SQL Analyzer takes SQL you already have. The OpenAPI Spec reader starts from an API specification. The Architecture Copilot works at the architecture level, and simulation has its own copilot for that context.
AI ▸ AI Model AssistantAI ▸ SQL AnalyzerAI ▸ OpenAPI SpecAI ▸ Architecture CopilotSimulate ▸ Simulation Copilot
Review before you accept
Output is presented for review rather than applied on your behalf. Read it as a draft from a fast colleague who has not met your business — the structure is often right and the semantics need your judgement.
Put the draft through the same checks as anything else
Run Validate Model and the naming check over an accepted draft. A generated model is not exempt from your standards, and this is the fastest way to find where it diverged from them.
Data ▸ Validate ModelData ▸ Naming Standards
What happens next
The reviewed draft is an ordinary model — it generates DDL, participates in lineage and impact analysis, and is versioned like anything else.
Example
An OpenAPI specification read into a first-pass model: the entities and relationships were largely right, and the two hours saved went into the naming and definitions that the specification could not have known about.
Tips
- Give it the real material rather than a description of it. A specification produces a better draft than a paragraph about the specification.
- Review the relationships hardest. Attributes are easy to check by eye; a wrong cardinality is not, and it is the one that costs later.
Limitations
- Assistants propose; they do not decide. Nothing is applied to your model without your review.
- AI capabilities are separately enabled.