AI document translation
Automatic translation of policies and contracts into 5 languages preserving layout, tables and headers.
What is the AI Document Translation module?
AI translation is the module that automatically translates insurance policies, contracts and technical documents across 5 European languages (Italian, English, German, French, Spanish) fully preserving the original document layout: tables, headers, footers, font formatting, field positioning. The base model is fine-tuned on multilingual insurance corpora and applies a vertical-sector glossary (fideiussione → surety bond / Bürgschaft; escussione → payout; appendice → endorsement) to guarantee terminological consistency. On-tenant pipeline for sensitive data, integration with policy workflow, optional human-in-the-loop for high-value documents. Replaces external translation agencies (3-5 day turnaround, significant per-page cost) with a 1-10 minute per-document workflow.
Who has multilingual portfolios
What the AI translation module does
- 5 languages: IT, EN, DE, FR, ES (all combinations)
- PDF and DOCX layout preservation
- Tables, headers, footers intact
- Font adaptation to new text size
- Built-in insurance terminology glossary
- Custom customer glossary configurable
- On-tenant pipeline for sensitive documents
- Certified enterprise cloud provider (EU DPA)
- No calls to public LLMs
- Optional HITL workflow for high-value docs
- Linked multilingual version history
- Side-by-side IT/EN diff for review
From original document to multilingual
Document upload
PDF or DOCX uploaded manually or received via PEC. System automatically identifies source language.
Text + layout extraction
OCR (if needed) + structured text extraction with bounding boxes. Tables, headers, sections mapping preserved.
AI translation
Enterprise LLM applies translation + insurance glossary + customer glossary. Output: translated text with same semantic structure.
PDF reconstruction
System replaces original text with translated text in the same positions. Font adapted if new language needs more/less space.
Optional review (HITL)
For high-value documents: human reviewer sees side-by-side IT/EN diff, accepts or corrects. Corrections feed glossary fine-tuning.
Archival & linking
Translated version archived and linked to original as \"translation of\". Full audit trail. Version available via portal and API.
Technical stack
Impact vs traditional translation
Frequently asked questions about AI translation
Which languages are supported?
5 main languages of the European insurance market: Italian (IT), English (EN), German (DE), French (FR), Spanish (ES). All bidirectional combinations are supported (e.g. IT → EN, EN → DE, DE → IT). For extra-EU languages (Chinese, Arabic, Russian) an extension is available on request with dedicated setup timing.
How is the document layout preserved?
The module works at three levels: (1) text extraction with bounding box on the original page; (2) translation of text only, leaving tables, headers, footers, formatting intact; (3) PDF reconstruction with translated text in original positions, adapting fonts to new text size. Result: translated PDF visually indistinguishable from the original except for the language.
Is insurance terminology translation accurate?
Yes. The base model is fine-tuned on multilingual insurance-document corpora, with a vertical-sector glossary (e.g. "fideiussione" → "surety bond" / "Bürgschaft" / "caution" / "fianza"; "escussione" → "claim payout" / "Inanspruchnahme"; "appendice" → "endorsement" / "Nachtrag"). Glossary configurable by customer for internal or niche-specific terms.
Is human review available?
Yes, optional and configurable. For high-value documents (e.g. high-premium policies, binding-authority contracts) you can activate the human-in-the-loop workflow: AI translation is a first draft, a human reviewer (internal customer or certified partner) approves or corrects before release. Corrections feed glossary fine-tuning.
Are sensitive data sent to an external LLM?
Depends on customer-chosen configuration. (A) Self-hosted: enterprise LLM model runs on-tenant, zero external data — recommended option for sensitive Italian policies. (B) Certified cloud provider: translation via enterprise API (e.g. Azure OpenAI, Anthropic) with DPA forbidding secondary use of data and EU residency. Never calls to public LLMs (ChatGPT, Claude.ai).
How long does a typical translation take?
Standard 5-10 page policy: 1-2 minutes for automatic translation + human review time (10-20 minutes if activated). More complex contracts (binding authority 30-50 pages): 5-10 minutes automatic + 1-2 hours human review. Comparison: external translation agency 3-5 business days + significant per-unit cost.
Continue exploring the platform
Want to see AI document translation in action on your real flows?
45 minutes with one of our engineers, no sales script. You show us your current process and we show you concretely how this module would solve the critical points.