Case 03 · Insurance · AI
Documents read, checkedand routed in the background.
An onboarding pipeline for an insurer that classifies customer documents, extracts and cross-checks the data, and sends only uncertain cases to a person — with the evidence highlighted.
- Industry
- Insurance
- Region
- Europe
- Engagement
- Product Team · Discovery to launch
- Services
- AI & ML · Product Engineering
The challenge
An insurer’s onboarding depended on people reading documents: IDs, proof of address, vehicle registrations, bank statements. Every application waited in a queue for manual checks, errors slipped through, and spikes in demand turned into backlogs that lasted weeks.
What we built
- A document pipeline that classifies each upload, extracts the fields that matter and checks them against the application and against each other.
- Quality checks at upload — blur, glare, cropping and expired documents are caught while the customer is still there to fix them.
- Confidence scoring: clear cases pass automatically; uncertain fields go to a reviewer with the relevant part of the document highlighted.
- A review workspace for the operations team, with keyboard-first flows and a full audit trail.
- Integration with the policy administration system and the KYC provider.
How we did it
We started with a labeled sample of real, anonymized documents to measure accuracy field by field. OCR and vision-language models do the reading; checks that must be deterministic stay as rules.
- Deployed in the client’s private cloud, so documents never left their infrastructure
- Confidence thresholds set together with the compliance team
- Rollout one document type at a time
The outcome
- ✓Routine documents processed without manual data entry
- ✓Reviewers focus on edge cases, with the evidence in front of them
- ✓Problems caught at upload instead of days later
- ✓Every automated decision traceable for audit
Stack
- Python
- PyTorch
- Vision-language models
- OCR
- FastAPI
- PostgreSQL
- React
- Kubernetes
- Private cloud
[ Contact ]
Have a complex product in mind?Let’s talk.
> start_project --team neurovista
✓ reply within one business day · NDA on request
or write to {{CONTACT_EMAIL}}