Star schema

AI in the Star schema designer

The Star schema designer leans on AI more than any other part of Kenseme. AI proposes the design and drafts the SQL; you review every step. Every AI step has a manual alternative.

Where AI runs

When What AI does Manual alternative
Plan setup, Run AI Analysis Reads the source schema (split by subject area when it is large) and proposes dimensions, facts, dimension lookups, grain, and build order, guided by your AI Guidance. Start with empty plan, then + Add Item Manually
Review, Refine selected with guidance… Redesigns only the ticked items using your extra guidance. Untick the item and add it by hand later
Step A, source view Drafts the CREATE VIEW that combines several source tables, from the item’s design instruction. Regenerate with AI takes extra guidance. Write SQL myself or Pick a single existing table instead
Build Pipeline Drafts source views for items that need one, as part of the background build. Build items one at a time
Refresh Prompts Rewrites an item’s design instructions after its source changed or a dimension it used was dropped. Edit the view or procedure yourself
Version scripts, Validate script with AI agent Reviews generated SQL for dialect problems. Leave the box unticked

Tables (Step B), load procedures (Step C), the date dimension, and the master procedure are generated from the plan’s settings and lookups, not written by AI.

What AI is good at

  • Spotting dimension candidates: lookup tables and attributes that many tables share.
  • Spotting facts: transactional and event tables, with a sensible grain.
  • Proposing business keys from names and keys in the source.
  • Suggesting which dimensions need history (SCD2).
  • Gathering scattered flags into a junk dimension.

Where to check its work

  • Role-playing dimensions. Check that DimDate (OrderDate) and DimDate (ShipDate) point at the right source columns.
  • Volumes. AI sees structure, not row counts. Tell it about very large or very small tables in AI Guidance.
  • Your conventions. Naming comes from the plan’s settings, but column choices and grain are the AI’s. Read each item’s reasoning in the review step.
  • Source views. Read every drafted view before Create View & Next >. Nothing is created until you click it.

What you control

  • Whether AI runs at all.
  • Which proposed items become part of the plan.
  • The guidance AI works from, at setup, on refine, and when regenerating a view.
  • Every view before it is created, and every lookup the load procedures use.

What AI does not do

  • It doesn’t create anything without a click from you (or a Build Pipeline you started).
  • It doesn’t read your data. It works from the schema’s structure and your guidance.
  • It doesn’t deploy anything to your database.

Cost and limits

Every AI step counts toward your organization’s AI usage. When the allowance is used up, new AI analyses and builds don’t start. See AI token allowance and usage.