Technology
Technology OverviewImage Trust LayerDocument Trust LayerAudio Trust LayerInformation Trust LayerSource VerificationProducts
Fraud ScannerSafeCamSafeMicGardioIndustries
InsuranceBanking & KYCAuthoritiesMedia & PressHR & ExpensesDefenseResources
BlogCase StudiesAPI DocsPress AreaCompany
About UsJobsContactThe Document Trust Layer is VAARHAFT's proprietary deep learning architecture for detecting forged, manipulated, and AI-generated documents. Trained on real-world fraud data, not academic benchmarks.
Each module handles a specific document analysis task. Proprietary neural networks for forgery detection, deterministic algorithms for structural validation, and deep metadata extraction. All accessible through a single endpoint.
Detects fully AI-generated documents created by GPT-4o, Gemini, or other generative models. Identifies synthetic layouts, fonts, and content patterns.
Detects post-creation document forgery: altered amounts, swapped fields, modified signatures, copy-paste artifacts, and font rendering inconsistencies.
Internal and external duplicate detection across document repositories. Identifies resubmitted, recycled, or cloned documents beyond simple hash comparison.
Extracts and cross-references PDF metadata, creation tools, font embedding, compression artifacts, and edit history to verify document authenticity.
Inspects the internal structure of PDF files: object trees, stream encoding, font subsetting, and layer composition to detect structural anomalies.
Automatically classifies document types (invoices, contracts, ID documents, certificates) to route each document to the appropriate verification pipeline.
No off-the-shelf foundation models. Our entire document analysis pipeline is built from scratch with specialized architectures for structural, visual, and semantic document analysis, trained exclusively on proprietary forensic datasets.
Multi-Layer Document Analysis
Analyzes documents at three levels simultaneously: visual appearance, structural layout, and semantic content consistency.
Intrinsic Explainability
Explainability is an inherent mathematical part of backpropagation, not generated post-hoc like GradCAM or ELA. The proof is part of the classification.
Deterministic, No LLM
100% reproducible results. No stochastic LLM, no prompt engineering, no hallucinations.
Adversarial Hardening
Hardened against adversarial perturbations and real-world document laundering: scans, faxes, re-exports, and multi-generation PDF compression.
A highly optimized, multi-stage feature extraction pipeline trained to detect document-specific forgery artifacts, even after scanning, faxing, or re-exporting. Intrinsic explainability as a mathematical byproduct, no post-hoc XAI.
Academic benchmarks test on pristine, high-resolution PDFs. Fraudsters use scanned copies, faxed reproductions, and re-exported files. The benchmarks from most document forensics providers are worthless.
Our model was not trained in a lab but on real-world document fraud data. We evaluate our metrics explicitly on degraded documents: scans, faxes, and re-compressed PDFs.
80% Precision @ 1% Prevalence: Even when 99 out of 100 documents are authentic, our model is correct in 8 out of 10 cases when flagging a document as forged.
Our performance holds even when a forged document was:
FP-Catcher Module
Pure document analysis is not enough for production. Specialized false-positive catcher modules are adapted to real production conditions. Example: Watermarks and stamps are automatically detected and filtered to prevent false triggers on legitimate overlays.
Different use cases require different trade-offs. Automated rejection needs maximum precision. Screening with human-in-the-loop maximizes recall.
Choose the mode that matches your risk appetite, or let us calibrate your model individually.
Balanced: Balanced trade-off between precision and recall. Default configuration for most enterprise customers.
Our out-of-the-box metrics are industry-leading, but every company has unique document distributions. We offer deep system calibration to tune the model precisely to our enterprise customers' requirements.
Through advanced probability calibration and model finetuning, we ensure that a confidence score of 90% truly corresponds to a 90% probability of forgery. This makes it optimal for seamless integration into existing risk engines.
Our model was not trained in an academic lab. Over 5 million in-house curated document samples, hand-manipulated by forensic experts, simulate extreme real-world conditions including scanned copies, fax degradation, and targeted field manipulation.
Most open-source document forensic tools rely on template matching or simple metadata checks. They fail on scanned copies, faxed reproductions, and re-exported PDFs because the visual artifacts they depend on are destroyed in the process. Our proprietary models are trained specifically on these degraded inputs, which is why our accuracy holds in conditions where conventional approaches become unusable.
Zero-Day GenAI Defense
When new AI document generators are released, we integrate their signatures within days, not weeks.
GPU clusters (NVIDIA Tensor Core) scale dynamically. 10 requests at night or 5,000 RPM during a campaign, and latency remains stable.
Specialized sub-models vote independently on generation and manipulation probability, combined via calibrated ensemble.
Dedicated endpoints with no serverless cold-start penalty. Inference optimized via TensorRT (INT8 Quantization).
Average end-to-end latency including ensemble voting, heatmap generation, and JSON response.
Zero-retention API: We don't store any documents. RAM is cleared immediately after inference. Privacy by design.
Supports all common document formats natively: PDF, TIFF, JPEG, PNG, WebP. Multi-page PDFs processed page by page.
Direct upload up to 10 MB for single documents. For large payloads and batch pipelines, asynchronous S3 offloading up to 80 MB with webhook notification on completion.
Over 5 million in-house curated document samples simulating extreme real-world conditions including scan degradation and targeted field manipulation. Robustness where conventional SOTA models break down.
When new AI document generators are released, we integrate their signatures within days. Radical agility in the cat-and-mouse game of AI-generated fraud.
Developed and hosted in Germany. Zero-retention API: RAM is cleared immediately after inference, no customer data stored. Fully auditable, EU AI Act compliant, no third-party data leakage.
Our architecture generates explainability not after the fact, but as a mandatory mathematical byproduct. Granular JSON data and spatial heatmaps prove pixel-precisely why a document was flagged.
Built from the ground up for scale. Asynchronous webhooks, deterministic processing without LLM overhead, and seamless S3 payload offloading for gigabyte-heavy document pipelines.
Specialized modules for real production conditions. Watermark detection, stamp filtering, and adaptive threshold adjustment minimize false positives.
The Document Trust Layer is the core document detection engine of the VAARHAFT Fraud Scanner, available as a fully managed API. No infrastructure, no model maintenance. Integrate once, benefit permanently.
{
"suspicion_level": "Red",
"Files": {
"invoice.pdf": { … },
"contract.pdf": { … }
},
"caseNumber": "Case 456B",
"sessionId": "7c2e9af1-5512-48b3-a1d7-ee43bc8d91f0",
"modelVersions": {
"imagesModelVersion": "vh-mo-images-alpha",
"documentsModelVersion": "vh-mo-docs-alpha"
},
"tokensConsumed": 18,
"attachments": {
"heatmaps": [ … ],
"analysis_report": [ … ]
}
}
We detect AI-generated documents from systems like GPT-4o and Gemini, post-creation manipulations such as altered amounts or swapped fields, duplicate submissions across repositories, and structural anomalies in PDF files. Our six modules cover the full spectrum from synthetic document generation to targeted field-level tampering.
Unlike tools that rely on clean digital PDFs, our models are trained explicitly on degraded inputs: scanned copies, smartphone photos, faxed reproductions, and re-compressed files. Detection accuracy remains stable because our training data mirrors real-world conditions, not lab-quality originals.
Yes. Our tampering detection module identifies post-creation manipulations at the field level: altered amounts, modified signatures, swapped fields, copy-paste artifacts, and font rendering inconsistencies. The heatmap output shows exactly which document regions triggered detection.
We deploy specialized FP-Catcher modules for production conditions. Stamps, corporate logos, and existing watermarks are automatically detected and excluded from the forensic analysis. Enterprise customers can additionally configure custom thresholds for their specific document types.
The API supports PDF (single and multi-page), TIFF, JPEG, PNG, and WebP natively. Multi-page PDFs are processed page by page. For large batch pipelines, asynchronous S3 offloading handles files up to 80 MB with webhook notification on completion.
No. The Document Trust Layer uses a zero-retention architecture. Documents are processed in GPU RAM and cleared immediately after inference. No customer documents are stored, logged, or retained. The API is fully stateless, 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.
Detects forged, manipulated, and AI-generated documents.
Book a demo and see how our detection performs on your real-world document pipeline. We'll walk you through integration, threshold configuration, and deployment options.
