Technology
Technology OverviewImage Trust LayerDocument Trust LayerAudio Trust LayerInformation Trust LayerSource VerificationProducts
Fraud ScannerSafeCamSafeMicGardioIndustries
InsuranceBanking & KYCAuthoritiesMedia & PressHR & ExpensesDefenseResources
BlogCase StudiesAPI DocsPricingPress AreaCompany
About UsJobsContactThe Information Trust Layer is VAARHAFT's contextual intelligence layer that goes beyond visual forensics: agentic verification, geolocation prediction, structured data extraction, and content moderation. Autonomous plausibility checks catch what pixel analysis alone cannot.
Each module adds a layer of contextual understanding beyond visual forensics. Proprietary models for content classification, deterministic algorithms for data extraction, and agentic AI for autonomous plausibility verification. All accessible through a single endpoint.
Predicts the geographic origin of an image from pixel content alone, no metadata required. Proprietary model trained on global geospatial features for location plausibility verification.
Autonomous AI agents verify document claims against external sources in real time. OSINT-driven background verification of companies, addresses, and financial data without human intervention.
Extracts structured data from documents: invoice amounts, dates, addresses, tax IDs, and line items. Deterministic extraction pipeline.
Identifies depictions of minors in uploaded media for content safety compliance. Proprietary classification model for age-group estimation in visual content.
Detects nudity and explicit content in uploaded images for automated content moderation. Proprietary model with configurable severity thresholds.
Detects and decodes embedded QR codes, URLs, and hyperlinks within documents and images. Identifies potential phishing links, malicious redirects, and suspicious embedded targets.
Extracts and validates phone numbers from documents and images. Deterministic PII detection for compliance workflows, contact verification, and fraud pattern analysis.
Visual forensics detects what was changed. The Information Trust Layer understands whether what is claimed makes sense. An agentic verification layer that autonomously cross-references document content against real-world data sources.
Contextual Plausibility
Goes beyond pixel-level analysis to verify whether the content of a document is internally consistent and externally plausible.
Agentic Verification
Autonomous AI agents execute multi-step verification workflows, querying external registries, maps, and databases without human prompting.
Structured Data Extraction
Deterministic OCR pipelines extract structured fields from documents, enabling automated validation of amounts, dates, and identifiers.
Content Safety Classification
Proprietary models for detecting minors, nudity, and embedded threats. Content moderation as an integrated layer, not a separate service.
A multi-stage orchestration pipeline that combines deterministic extraction, proprietary classification models, and agentic verification workflows. Each module contributes structured signals that are aggregated into a unified trust assessment.
A perfectly forged document can pass every visual forensic test. The Information Trust Layer adds a second dimension: autonomous agents that verify whether the claims inside a document are true, not just whether the document looks authentic.
Each agentic check executes a multi-step verification workflow against external data sources in real time. No human intervention, no manual OSINT. The system decides what to verify and how.
Verifies that the issuing company exists, is active, and matches the claimed registration number against official commercial registries.
Cross-references stated addresses against Google Maps and postal databases. Validates that the address exists and is plausible for the document type.
Recalculates line items, tax rates, and totals from extracted OCR data. Detects arithmetic inconsistencies that indicate manual manipulation of amounts.
Compares billed amounts against known fee schedules and rate tables. Identifies charges that deviate from standard pricing for the claimed service.
Compares claims across multiple submitted documents. Detects contradictions in dates, amounts, addresses, or identifiers across a document set.
Contextual verification cannot rely on clean, structured inputs. Real-world documents arrive as scans, photos, compressed PDFs, and blurry smartphone captures. Our extraction and verification pipeline is hardened for these conditions.
Agentic checks execute against live external sources (registries, maps, databases) and handle timeouts, ambiguous results, and partial matches gracefully.
Full deployment on customer infrastructure. Docker containers, Kubernetes-ready, air-gapped operation. Agentic checks route through customer-controlled egress.
Single REST endpoint for all seven modules. Select modules per request, receive structured JSON with per-module confidence scores and verification results.
Autonomous agent runtime manages multi-step verification workflows. Parallel execution, retry logic, and graceful degradation when external sources are unavailable.
End-to-end processing including OCR extraction, agentic verification against external sources, and aggregated JSON response.
Zero-retention architecture: documents are processed in memory and immediately discarded after analysis. No customer data is stored, logged, or retained.
Agentic checks connect to commercial registries, geolocation services, postal databases, and fee schedule repositories. Configurable source priority per customer.
Visual forensics catches manipulated pixels. The Information Trust Layer catches manipulated meaning. Autonomous verification of whether document claims are true, not just whether they look authentic.
No static rule engines. Autonomous agents decide what to verify and how, adapting to each document type. Multi-step reasoning against live external data sources without human intervention.
Geolocation, OCR, content moderation, and agentic verification, all through a single endpoint. No separate vendor integrations, no data duplication across services.
Developed and hosted in Germany. Zero-retention architecture: all data is processed in memory and cleared immediately. Fully auditable, EU AI Act compliant, no third-party data leakage.
Extraction and verification pipelines tested on scanned copies, smartphone photos, compressed PDFs, and degraded documents. Production accuracy, not lab accuracy.
Designed to work alongside the Image Trust Layer and Document Trust Layer. Visual forensics plus contextual intelligence, two dimensions of trust in a single Fraud Scanner pipeline.
The Information Trust Layer is the contextual intelligence engine of the VAARHAFT Fraud Scanner, available as a fully managed API. Adds agentic verification, data extraction, and content moderation to visual forensic analysis.
{
"suspicion_level": "Yellow",
"Files": {
"invoice.pdf": {
"file_level_analyses": { "plausibility": {
"handelsregister_verification": { "names_match": true, … },
"page_number_consistency": { "is_consistent": true }
} },
"items": [ { "analyses": { "plausibility": {
"document_deep_analysis": { "math_correct": true, "goae_plausible": false },
"qr_code_analysis": { "qr_detected": true, … }
} } } ]
}
},
"cross_file_analyses": { … },
"agenticRecommendation": { … },
"tokensConsumed": 18
}
Visual forensics detects what was changed in the pixels. Contextual verification checks whether the claims inside a document make sense: whether the issuing company exists, the address is real, and the amounts add up. A perfectly forged document can pass every visual test. The Information Trust Layer catches it through its content.
Autonomous AI agents execute multi-step verification workflows against external data sources in real time: company registries, maps and postal databases, fee schedules, and cross-references across entire document sets. The system decides what to verify and how, without human intervention or manual OSINT research.
More than 10 external sources, including official commercial registries, geolocation and postal databases, and fee schedule repositories. Source priority is configurable per customer, and on-premise deployments route all checks through customer-controlled egress.
A proprietary model predicts the geographic origin of an image from pixel content alone, using global geospatial features. This enables location plausibility verification even when EXIF data is missing, stripped, or deliberately manipulated.
Yes. Proprietary models detect depictions of minors and nudity, and deterministic modules decode QR codes, links, and phone numbers to flag phishing attempts and PII. Content moderation runs as an integrated layer in the same API call, not as a separate service.
End-to-end latency is under 3 seconds including external source verification, with all checks executed in parallel. The architecture is zero-retention: documents are processed in memory and immediately discarded. Hosted in Germany and EU AI Act compliant.
Every VAARHAFT product is built on the same proprietary technology base: five specialized trust layers covering images, documents, audio, contextual information, and capture-time source verification. Combined, they form the technological foundation of the VAARHAFT Trust Suite.
Verifies document claims through agentic plausibility checks.
Book a demo and see how contextual verification catches what visual forensics misses. We'll walk you through agentic checks, OCR extraction, and deployment options.
