The problem
Documents are scattered, contract changes are difficult to spot, and conventional extraction hides uncertainty. A plausible AI answer can still be wrong — unacceptable for premiums or cancellation deadlines.
Advanced beta / pre-release
SaaS · InsurTech · Independent project
Insurana
Insurance data that never becomes a black box.
Insurana structures policies, links every critical field to evidence, and keeps people in control before AI output becomes trusted contract information.
Dashboard · 2026
Sample household
2 insured people
Home contents & personal liability Alpina Versicherung · POL-26-1842
2026Evidence coverage 92%
1 field requires review
Advanced beta / pre-release
The technical core and most of the feature set are implemented. Access remains deliberately limited while quality, privacy and operational readiness continue to be validated for broader use.
11 controlled pipeline steps
Multi-tenant role-based access layer
Automated backend, E2E & accessibility tests
CI/CD delivery concept with quality gates
Product thinking
Not more documents. More confidence.
Insurance policies contain deadlines, limits and exclusions that are rarely top of mind — until they matter. Insurana turns dense documents into reviewable, versioned information.
The solution
Every relevant value remains linked to its page, excerpt, model version, confidence and review state. Critical or changed fields pass a human review gate before influencing deadlines or active policy data.
Traceable
Every value links back to the exact location in the source document.
Confirmed
Uncertainty is made visible instead of hidden behind automation.
Versioned
Endorsements create revisions rather than silently overwriting confirmed data.
Responsibility
One product, built end to end by me.
I own Insurana from product concept to controlled operations: experience design, domain model, APIs, document processing, billing, security, testing and delivery.
01
Product & experience
- Product discovery and feature planning
- UI/UX, responsive interface and accessibility
- Review, consent and versioning workflows
02
Platform engineering
- SvelteKit frontend and modular FastAPI service
- PostgreSQL domain model and asynchronous jobs
- Document, evidence and policy logic
03
Trust & operations
- Authentication, authorization and audit trail
- Test strategy, CI/CD and monitoring
- Privacy, billing and controlled deletion
Product lab
Explore the core logic yourself.
Three interactive, fully synthetic examples show how Insurana processes documents, exposes evidence and compares policy versions.
Synthetic sample data. All names, values, documents and results in this demonstration are fictional. No real insurance or personal data is processed.
Document pipeline
From upload to an actionable deadline: every step has one clearly bounded responsibility.
Select a step. Navigate with arrow, Home and End keys.
Step 01 of 11
Upload
The document is assigned to an account and profile before any domain interpretation begins.
Signed upload · Ownership contextEvidence explorer
An extracted value becomes useful only when its source, uncertainty and review state are visible together.
Select a field. Navigate with Up/Down arrow, Home and End keys.
Review required Synthetic sample data
Cancellation deadline
30 September 2026
Confidence 72%
- Source
- Policy_2026.pdf
- Page
- 5
Evidence excerpt“Cancellation must reach us no later than three months before the policy expires.”
Confirmed Synthetic sample data
Gross negligence
Included up to CHF 5'000'000
Confidence 88%
- Source
- Additional_terms.pdf
- Page
- 7
Evidence excerpt“The defence of gross negligence is waived up to the insured amount of CHF 5'000'000.”
Policy diff
A new annual policy never silently overwrites the previous one. Relevant changes are shown field by field with their status.
| Field | Version 2025 | Version 2026 | Status |
|---|---|---|---|
| Annual premium | CHF 1'120.00 | CHF 1'248.00 | Changed +11.4% |
| Deductible | CHF 300.00 | CHF 500.00 | Changed +CHF 200 |
| Cyber protection | Not included | Included up to CHF 20'000 | Added New |
| Latest cancellation date | 30 September 2025 | 30 September 2026 | Changed New period |
Architecture
Modular, asynchronous and designed for revision safety.
Domain logic is separated from transport, storage and background processing. This limits side effects and makes critical flows directly testable.
01
Experience layer
SSR-capable interface for uploads, reviews, policies, deadlines, billing and administration.
SvelteKitTypeScriptTailwind CSSPlaywright
02
Domain & API
Modular domains with centralized ownership checks and explicit service boundaries.
FastAPIPydanticSQLAlchemyREST API
03
Data & jobs
Transactional revisions, controlled migrations and repeatable background work.
PostgreSQLAlembicRedisARQ
04
Runtime & storage
Designed for separate staging and production environments with object storage and a TLS proxy.
DockerCaddyHetzner CloudMinIO / S3
Security & privacy
Trust is an architectural decision.
Insurance documents can contain highly sensitive data. The protection concept is therefore designed to enforce authorization boundaries in server-side data operations, not only in the interface.
Tenant isolation
The protection concept uses centralized ownership checks and role-based authorization boundaries.
Secure sessions
The session concept accounts for OAuth hardening, token rotation and misuse detection.
Protected files
The file-protection concept combines signed access, file checks, malware scanning and controlled deletion.
Privacy by design
The concept includes data minimization for AI providers, server-side consent and redacted error tracking.
Administrative control
The administration concept includes MFA, time-limited support grants and audit events.
Supply chain
The delivery concept provides for secret, dependency and container scans.
Quality & operations
Built for change, not merely built.
Automated tests, safe migrations and observable releases are part of the product rather than a later addition.
Delivery pipeline
- 01 Lint & frontend check
- 02 Backend & E2E tests
- 03 Migration test
- 04 Security scans
- 05 Container build
- 06 Deploy by commit SHA
- 07 Readiness & smoke tests
- 08 Rollback concept & incident runbooks
Test strategy
Automated backend tests cover more than happy paths, with particular focus on tenant isolation, concurrent endorsements and sensitive lifecycles.
- Authentication, RBAC and authorization boundaries
- Document upload, file and malware checks
- Policies, review and concurrent revisions
- Billing, consent, sharing and account deletion
- E2E and accessibility flows with Playwright
External systems
- Payments
- Stripe
- Resend
- OAuth
- Monitoring
- Sentry
- Storage
- MinIO / S3
- CI/CD
- GitHub Actions
Status & roadmap
Advanced beta, deliberately not a launch promise.
The product core works. Before broad market entry, assumptions will be validated against realistic anonymized references and independent reviews.
01 Implemented
Technical core
The essential product and operational building blocks are in place.
- Policy, document and review lifecycle
- Versioning, deadlines, billing and audit
- Staging plus CI/CD, monitoring and backup concepts
02 In validation
Launch readiness
Quality and resilience are being demonstrated systematically.
- AI quality with anonymized reference documents
- Operational readiness under realistic conditions
- Independent review of privacy and product boundaries
03 Afterwards
Controlled release
Access will expand gradually and observably.
- Limited beta cohort and feedback cycles
- Verify operational runbooks under real use
- Production approval against defined quality gates
Engineering decisions
The hardest problems lived between the layers.
AI without an illusion of authority
Confidence alone is insufficient. Evidence, field criticality and human confirmation form the trust model together.
Revisions without data loss
Optimistic concurrency control and transactional locks protect confirmed data from late or parallel analyses.
Deadlines as derived truth
Reminders originate only from confirmed fields and stay linked to the policy revision that created them.
Project conversation
Interested in the architecture behind Insurana?
I am happy to discuss product decisions, security boundaries or the document pipeline — without disclosing sensitive implementation details.