Skip to main content

Building a Transformation

This is where Visitran AI truly shines. Using the Transform intent, you can describe a data model in plain English and the AI will generate a complete transformation — with joins, aggregations, filters, and calculated columns — then materialize it directly to your database.

Setting Up​

Switch the intent selector to Transform before typing your prompt. If you're following the onboarding guide, click "Try This" on the transform task — it will set the intent and auto-fill the prompt for you.

Example Transform Prompts​

Here are some transformations you can try with the Jaffle Shop dataset:

Simple join:

"Create a model that joins orders with payments to show order totals"

Aggregation:

"Build a customer summary model showing each customer's total orders, total spend, and first order date"

Multi-step pipeline:

"Create a staging model that combines customers, orders, and payments into a single denormalized view"

Sending the Transform Prompt​

Type your prompt and click Send. The AI will begin working through multiple stages:

Stage 1: Planning​

The AI analyzes your request, examines the available tables and their schemas, and designs a transformation plan.

You'll see the thought chain showing:

  • Which tables the AI selected as sources
  • What joins it plans to use
  • What columns and calculations it will create

Stage 2: Generation​

The AI generates the model configuration — a YAML definition that describes the transformation in Visitran's no-code format.

Stage 3: Review & Apply​

Once generation is complete, you'll see an "Apply" button. This is your chance to review before materializing.

Click Apply to materialize the model.

Model Materialization​

After clicking Apply, Visitran materializes the model to your database. You'll see a progress indicator showing the materialization status:

Once complete:

  1. The new model appears in the file explorer on the left
  2. You can click to open it in the no-code editor
  3. The data is now in your database, ready to query

Inspecting the Generated Model​

Click on the new model in the file explorer to open it. You'll see:

  • Column list — All columns the AI created, including calculated fields
  • Transformation steps — The joins, filters, and aggregations applied
  • Generated SQL — The SQL query that Visitran compiled from the model configuration

This is the key advantage of Visitran's approach: the AI doesn't generate opaque code. It generates a model that flows through the same no-code engine, giving you full visibility and editability.

Fine-Tuning with No-Code​

After the AI generates a model, you can refine it using the no-code UI:

  • Add or remove columns using the column panel
  • Modify joins in the transformation configuration
  • Add formula columns for custom calculations
  • Change aggregations or add filters

This hybrid workflow — AI generation + manual fine-tuning — gives you the speed of AI with the precision of manual control.

tip

If the AI-generated model isn't quite right, you can start a follow-up conversation to refine it. Just describe what you want to change: "Add a payment_method column to the model you just created"

Next, let's explore the SQL intent for quick ad-hoc queries.