Database schema design

Tables that make bad data impossible

Your AI designs or reviews a relational schema the way a database engineer on call would: entities and keys, types and constraints that enforce meaning, tenancy and soft-delete decisions, indexes derived from your real queries, and runnable migration SQL.

By Vander Labs. A Vander Labs Original, written by the Vander Labs team with AI assistance. Works with Claude, ChatGPT, Cursor and Claude Code.

What your AI learns

Models business facts, normalizes to 3NF and denormalizes only on purpose
Chooses keys, types and constraints that encode the rules
Indexes from the actual queries and writes expand-migrate-contract migrations

How it works

Model entities: Entities with definitions and natural identifiers, relationships with cardinality, a Mermaid ERD.
Normalize deliberately: Third normal form, then denormalization only for a measured read path with an invalidation rule.
Choose keys: Identity or UUID surrogates, natural keys as unique constraints, foreign keys with explicit delete behavior.
Pick types: timestamptz, numeric for money, text with checks, lookup tables over enums, jsonb only for schemaless data.
Constraints: Not null by default, checks for ranges and states, unique and exclusion constraints for business rules.
Standard patterns: Timestamps, tenant_id first in indexes with row-level security, partial unique indexes under soft delete.
Index from queries: Composite indexes in equality, range, sort order per listed query, with explain analyze to confirm.
Write the migration: Ordered SQL files, concurrent index creation, destructive changes split so old code keeps running.
Review mode: Existing schemas checked against the same rules with severity and fix SQL.

Limits and safety

Says when a document store or a warehouse schema fits better than a relational design.
Refuses float for money and timestamp without time zone, and explains why.

Topics: postgresql, schema, sql, migrations, indexing