Step 2: Browse the AI-classified archive — Paperless-ngx Walkthrough

See how Amazon Nova Pro has tagged, titled, classified and extracted correspondents

Step 2 of 4 • 5 minutes

Step 2 5 minutes

Browse the AI-classified archive

Open the Documents view and see how Bedrock has classified every document — titles, tags, types, correspondents, summaries.

Every document arrives with an AI-rewritten title, AI tags, AI document type and AI-extracted correspondent — all generated by Amazon Nova Pro at consume time

Tags were created automatically as Bedrock classified each document

Every document is classified into one of seven types — created on first use by Bedrock

Senders and issuing bodies were pulled out of the OCR'd text and linked to each document

Expected outcome

• ✓ Documents tab shows ~36 sample parish documents with AI-rewritten titles
• ✓ Tags page shows AI-generated tags (planning, invoice, minutes, audit, …)
• ✓ Document types page shows seven AI-assigned types
• ✓ Correspondents page lists the senders Bedrock pulled out of the OCR'd text

What to look for

1. Open the Documents view

   Click Documents in the left nav. You should see ~36 cards, each one a sample parish document — planning notices, invoices, council minutes, agendas, correspondence, audit returns. Notice the title on each card: it's not the original filename like scan_24-01234-FUL.pdf, it's a human-readable summary like "Planning Notice 24/01234/FUL: Single-storey rear extension at The Old Forge". That's Amazon Nova Pro running on every consume.

2. Look at the tags

   Each card shows a few coloured tag chips. Click Tags in the left nav to see the full taxonomy that Bedrock built: planning, invoice, minutes, agenda, correspondence, audit, asset register, and so on. The tag count column shows how many documents Bedrock assigned each tag to.

   This taxonomy was not pre-loaded — it grew organically as each document was classified. New documents either join an existing tag or trigger a new one.

3. Look at document types

   Click Document Types. You'll see seven types — Letter, Invoice, Minutes, Planning Notice, Agenda, Report, Receipt — with counts. Bedrock chose one type per document.

4. Look at correspondents

   Click Correspondents. The list of senders has been pulled out of each document's OCR'd text and linked back to the originating document. You'll see Anglia Water Services Limited, Dunbar Grounds Maintenance Limited, Highways East — Footways Team, Information Commissioner's Office, Meadowbrook Parish Council, and so on.

   Each one was extracted from free text and turned into a structured record by Nova Pro on first sight.

5. Filter the documents

   Back on the Documents view, click a tag chip to filter — invoice shows you all eight invoices, planning brings up the dozen planning notices. Multiple filters compose; this is the raw search surface that the chat in Step 4 sits on top of.

What you're seeing: upstream Paperless-ngx has the structure for tags, document types, correspondents and titles, but populates them by hand or with a brittle pattern-matching classifier. This scenario plugs Amazon Nova Pro into the post-consume hook so every document is classified by a model that has actually read it. The cost per document is fractions of a cent.

Need help?

If you're stuck, check the troubleshooting section or contact us.

Press Escape to close