Currently verified
- Guided Telegram intake with a structured case
- Photo and document evidence associated with a case
- Conditional liability review
- Trade business selection by authorised people
- Verified roles for the Google Cloud services named here
Architecture for technical evaluators
The public Google Cloud flow behind a maintenance case.
The architecture connects language-based intake, structured case data, document processing, translation and human decision points—without disclosing the internal topology.
Public evaluator view
Product flow, not a system blueprint.
This page shows logical roles, selected Google Cloud services and scale drivers. Implementation, security and internal operating details deliberately stay outside the public view.
Simplified product flow
From a free-form report to human-decided follow-through.
The flow describes product responsibility, not technical call order or trust boundaries.
- 01
Free-form report
The tenant describes the problem in their own words and can add a photo, entry permission and availability.
- Telegram
- WhatsApp · pilot launch
- 02
Guided preparation
The workflow confirms category and problem detail and asks bounded questions for missing context.
- Gemini on Vertex AI
- Cloud Run
- 03
Structured case
Case state, description, media evidence and available time windows are brought together.
- Firestore
- Cloud Storage
- 04
Add context
Document fields and required ticket translations add to the available case context.
- Document AI
- Cloud Translation
- 05
Human decision
Administration reviews data and recommendations and selects a trade business or handles an exception manually.
- Administration
Text alternative to the product flow
- 01 · Free-form report: A free-form tenant report opens the case. Photos remain case evidence and are not presented here as AI-analysed input.
- 02 · Guided preparation: Gemini on Vertex AI supports language understanding and structured outputs; Cloud Run carries the guided service.
- 03 · Structured case: Firestore holds structured case state; Cloud Storage holds photo and document evidence.
- 04 · Add context: Document AI supports document extraction; Cloud Translation provides ticket text in supported target languages.
- 05 · Human decision: Administration remains responsible for review, required approvals and trade business selection.
Selected Google Cloud roles
Show only the services needed to explain the public product flow.
Gemini on Vertex AI supports language understanding and structured outputs. Cloud Run carries the guided service; Firestore and Cloud Storage hold case state and evidence; Document AI processes documents; and Cloud Translation provides ticket text in the required languages.
Generative AI
Gemini on Vertex AI
Supports language understanding, quality checks and structured outputs across varying text inputs.
Runtime
Cloud Run
Carries the guided intake service as a running product component.
Case data
Firestore
Holds the structured state of the case and workflow.
Evidence
Cloud Storage
Holds photo and document evidence associated with a case.
Documents
Document AI
Supports extraction of structured fields from documents.
Languages
Cloud Translation
Provides ticket text in supported target languages.
This selection explains product roles. It is not a complete infrastructure inventory or a reconstructable system diagram.
Scale drivers
Capacity follows the actual workload—not only the number of units.
These dimensions provide the basis for pilot measurement and later capacity planning. Public numbers follow only after a defensible measurement period.
Drivers, not published results
- 01
Maintenance cases
Eligible case count together with duration and outcome state.
- 02
Conversation and model use
Conversation steps, model calls and input/output tokens per case under an agreed definition.
- 03
Case evidence
Photo and document count and volume. Photos are not reported as current AI image analysis.
- 04
Document pages
Pages processed in agreed import or document workflows.
- 05
Translation
Translated ticket text and target languages within the agreed pilot scope.
- 06
Human review points
Approvals, exceptions and assignments that deliberately remain with authorised people.
Production readiness as an evaluation criterion
What must be demonstrated before operating at greater scale.
These are working and review principles, not a certification or blanket security assurance.
Review principles
- 01
Traceable operation
Important workflow states, errors and exceptions need to be observable and evaluable.
- 02
Measurable consumption
Model, document, storage and runtime use needs consistent definitions and reporting.
- 03
Clear decision boundaries
Recommendations, human approvals and manual exceptions must remain distinguishable in the product.
- 04
Reviewed data handling
Access, retention, export, deletion and regional requirements are reviewed service by service.
Status and evidence
Separate current behaviour, pilot prerequisites and later evidence.
Before pilot launch
- Implement and sign off WhatsApp technically
- Define measures and exception handling
- Operationalise the export and deletion procedure
- Confirm a secure route for required setup data
After sufficient measurement
- Publish pilot results with period and sample
- Publish capacity or cost claims only with reproducible definitions
- Prepare a private reviewer package only in response to a concrete request
Evaluate the technology on real cases in a pilot.
In an initial conversation, we clarify whether VerwaltKlar fits your portfolio, team and maintenance process.