Skip to main content

DAN AI — Custom Fields and Templates

DAN AI gives you three layers of control over what each extraction returns, from most local to most global:

  1. Per-document custom fields — tick a subset for a single extraction.
  2. Account-wide default fields — set the default field list per document type.
  3. Custom prompts — override the built-in prompt for a document type entirely.

You can mix all three. The lowest layer with an explicit value wins.

1. Per-document custom fields​

After upload, an optional Configure Fields step lets you tick the specific fields you want extracted from this document. If you skip it, DAN AI returns every field in the prompt's default schema.

  • Saved per-document in the custom_fields column.
  • Only affects this extraction — never changes the account-wide prompt.
  • Available from the dashboard upload modal and via the API's fields (or custom_fields) parameter — see REST API.

When to use it​

  • You only need a subset (e.g. just total_amount and vendor_name from a long multi-page invoice).
  • You're iterating: trying out a smaller field set to see if accuracy improves.
  • You want a one-off extraction tailored to an unusual file, without disturbing other users' defaults.

Example via API​

curl -X POST https://dan.sdlccorp.com/api/external/v1/extract \
  -H "X-API-Key: dan_live_xxxxxxxxxxxxxxxxxxxx" \
  -F "document=@invoice.pdf" \
  -F "doc_type=invoice" \
  -F "fields=invoice_number,total_amount,vendor_name,invoice_date"

The response's fields block will contain only those four keys (plus lines if line items were requested).

2. Account-wide default fields​

If your team always wants the same field list out of invoices (or any other type), set it once under Settings → Default Fields (admin only).

  • Pick a doc_type, then tick the fields that should be returned by default.
  • Save. Every future extraction of that type — dashboard, inbox sync, or API — uses this field list unless overridden.
  • A per-document or per-API-call custom_fields value still wins over the default.

Why use it​

  • Consistency — every team member's extractions return the same shape, making downstream code simpler.
  • Cost / latency — smaller field lists usually run faster and use fewer tokens.
  • Cleaner exports — Excel exports only show columns for the fields you care about.

3. Custom prompts (admin only)​

For non-standard layouts — invoices where the total is at the top, contracts in a non-English language, medical reports from a specific lab — you can replace the built-in prompt entirely.

Open Settings → Custom Prompts (admin only), pick a doc_type, and edit the prompt template. DAN AI uses your version for every extraction of that type from then on, across every ingestion path.

Iterating on a custom prompt​

tip

When iterating on a custom prompt, re-extract the same document multiple times to compare results without re-uploading. The raw AI response stays in document_extractions.raw_response for every run, so you can diff one attempt against the next.

A reasonable workflow:

  1. Save your current prompt as the baseline (download a copy).
  2. Make one small change.
  3. Re-run extraction on 3–5 representative documents.
  4. Compare the new responses against the baseline.
  5. Keep or revert.

When not to use a custom prompt​

  • One-off problems → fix with per-document custom fields instead.
  • General accuracy drift → check scan quality first, then doc_type selection. Prompts are powerful but easy to overfit.
  • Anything that mentions a specific company's proprietary identifiers — those belong in custom fields, not the prompt.

Layer precedence​

When all three layers are in play, DAN AI resolves them in this order:

  1. Per-request fields / custom_fields (from the dashboard modal or API call).
  2. Account-wide default fields for that doc_type.
  3. The prompt's default field set (built-in or custom).

If layer 1 is set, it wins outright. Otherwise layer 2. Otherwise layer 3.

Removing a custom prompt​

To revert to the built-in prompt for a doc_type, open Settings → Custom Prompts, select the type and click Restore Default. The change applies to the next extraction.