Fields & TablesSchema Management

Improving Schemas with AI

Use AI to automatically improve your field configuration

Improving Schemas with AI

You can use AI to automatically improve your Fields & Tables configuration. The AI will:

  • Correct data types (e.g. change "string" to "date" for date fields, "string" to "number" for amounts)
  • Improve descriptions to help extraction accuracy
  • Keep all field names and structure the same (only improves types and descriptions)

Where to improve

  1. Open your parser and go to Settings → Fields & Tables.
  2. Click Improve this schema using AI (top right, next to "Export Schema").
  3. The AI analyzes your schema and makes improvements.
  4. Review the changes and click Save Fields if you're happy with them.

What the AI improves

  • Data types: Changes field types to match the field name and purpose (e.g. "invoice_date" → Date, "total_amount" → Number)
  • Descriptions: Adds or improves descriptions to help the AI extract data more accurately
  • Field names: Not changed – all field names stay exactly the same
  • Structure: Not changed – tables stay as tables, fields stay as fields

What the AI does not change

  • Field names – All names stay exactly as they are
  • Field structure – Tables remain tables, fields remain fields
  • Required settings – Required field settings are preserved
  • AI extraction settings – Enable/disable AI extraction settings are preserved

After improving

  1. Review the changes the AI made (check field types and descriptions).
  2. If you're happy, click Save Fields.
  3. You may want to re-process existing documents so they use the improved schema (especially if data types changed).

Tips

  • Review changes: Always check what the AI changed before saving
  • Test first: Consider testing on a copy of your parser if you're unsure
  • Re-process: After improving, re-process a few documents to see if extraction is better
  • Use for new parsers: AI improvement is especially helpful when setting up a new parser

When to use AI improvement

  • New parser: When you've just created a parser and want to optimize it
  • Type corrections: When you notice fields have wrong types (e.g. dates as text, amounts as text)
  • Better descriptions: When you want more helpful descriptions for extraction

Next steps

How is this guide?

On this page