Claims Evidence Packager — Assessor-Ready Bundles from Photos, Invoices & Reports

Collate photos, invoices, incident reports and policy excerpts into a structured, assessor-ready evidence bundle — in milliseconds, not hours.

This is the check that moves upstream of the assessor: the completeness flag and the coverage cross-check surface a missing document or a policy-number mismatch before the bundle leaves your desk — not after an assessor bounces it back for rework. The pack arrives right the first time. Same claim in, same bundle out, every time.

---

What it does

Insurance claims handlers spend hours manually gathering scattered documents (emails, portals, scanned PDFs) and assembling them into a coherent evidence pack for assessors.
This Actor accepts a claim's metadata plus a list of document objects and returns a fully structured evidence bundle containing:

Section	What you get
Claim metadata	Normalised claimant, policy, and incident details
Document records	Classified, date-extracted, and checksummed document entries
Timeline	All documents ordered chronologically by detected date
Financial summary	Per-invoice breakdown and grand total claimed (AUD)
Coverage check	Cross-validates policy numbers mentioned in documents against the declared policy
Bundle summary	Document-type counts, completeness flag, and any documents with validation issues
Integrity checksums	SHA-256 per document — auditable chain of evidence

Document types recognised

photo, invoice/receipt, incident_report/police_report, policy_excerpt/policy_schedule, correspondence/email/letter, other

Date formats supported

• ISO 8601: 2024-03-15
• Australian: 15/03/2024
• Long form: 15 March 2024

---

Who buys this

• Insurance claims teams at Australian general insurers (motor, home, commercial)
• Loss adjusting firms that receive unstructured document bundles from claimants
• Claims management software vendors embedding evidence assembly into their workflow

---

Local demo

python3 main.py '{
  "claim_id": "CL-2024-001",
  "claimant_name": "Jane Smith",
  "policy_number": "POL-99001",
  "incident_date": "10/03/2024",
  "incident_description": "Vehicle damaged in hail storm on M1 motorway.",
  "documents": [
    {
      "id": "doc1",
      "type": "invoice",
      "content_text": "Invoice for panel repairs $3,450.00. Date: 2024-03-18. Policy POL-99001.",
      "source": "email",
      "filename": "panel_repairs_invoice.pdf"
    },
    {
      "id": "doc2",
      "type": "photo",
      "content_text": "Damage photograph taken 12 March 2024 at Goulburn St carpark.",
      "source": "portal",
      "filename": "damage_front.jpg"
    },
    {
      "id": "doc3",
      "type": "incident_report",
      "content_text": "Police event number E12345. Incident on 10/03/2024. POL-99001 holder present.",
      "source": "portal",
      "filename": "police_report.pdf"
    },
    {
      "id": "doc4",
      "type": "policy_excerpt",
      "content_text": "Policy schedule POL-99001. Comprehensive motor cover. Excess $750.",
      "source": "portal",
      "filename": "policy_schedule.pdf"
    }
  ]
}'

Run tests

$pytest tests/ -v

---

Pricing (pay-per-event)

Event	Price
Claim evidence bundle	USD $2.00 per claim processed

You pay per claim, not per seat. Manual collation typically takes 1–2 handler-hours per claim; at $2.00 a bundle, a 100-claim month costs USD $200 — no subscription, no unused capacity.

---

Input schema

See .actor/input_schema.json for full field definitions.

Output shape (abbreviated)

{
  "bundle_id": "A3F1...",
  "generated_at": "2024-03-20T04:12:00Z",
  "claim_metadata": { ... },
  "bundle_summary": {
    "total_documents": 4,
    "documents_by_type": { "invoice": 1, "photo": 1, "incident_report": 1, "policy_excerpt": 1 },
    "bundle_complete": true,
    "documents_with_issues": []
  },
  "timeline": [ ... ],
  "financial_summary": {
    "invoice_count": 1,
    "grand_total_claimed_aud": 3450.00
  },
  "coverage_check": {
    "declared_policy_number": "POL-99001",
    "policy_numbers_in_documents": ["POL-99001"],
    "consistent": true
  },
  "documents": [ ... ],
  "errors": []
}

---

FAQ

Does it read PDFs or scanned images? No — it takes document records with their text already extracted (content_text). You keep control of the extraction step; this Actor guarantees the assembly, classification, totalling, and cross-checking are consistent. It processes only the documents you supply — extraction, never acquisition: it never scrapes a portal, logs in on your behalf, or pulls from any interface no one offered you.

How are invoice totals calculated? Stated totals are detected so line items are never double-counted: an invoice reading "$1,200 + $1,650 + $2,000, Total $4,850" resolves to $4,850 — including subtotal + GST layouts. Amounts without a stated-total structure are simply summed.

Is this an AI tool? No. Deterministic parsing and cross-checking — the same claim always produces the same bundle, with SHA-256 checksums for an auditable evidence chain.

Does it decide claim outcomes? No. It organises evidence for human assessors. Coverage checks flag inconsistencies (e.g. a policy number mismatch); the decisions stay with your team.

Support & feedback

Open an issue on this Actor's Issues tab — it is monitored and responses are prompt.