Client-Side Biometric Processing
OCR, face detection, and liveness models run in the user's browser. Biometric data processing (local AI models) happens client-side and on our servers — raw frames are never sent to third-party servers.
How It Works
AgeEvidence uses local AI models to perform all biometric and document analysis directly in the user's browser:
Local AI models (Face Analysis)
- Face detection — a model for real-time face detection with quality assessment (centering, size, frontal pose)
- Face mesh — 500+ landmark mesh for precise facial geometry analysis
- Face embeddings — 1024-dimensional descriptor extraction for face matching between liveness and ID photo
- Anti-spoofing — detection of printed photos, video replay, and mask attacks
- Liveness detection — blink detection and head turn tracking for challenge-response verification
Document OCR
- Keyword detection — 100+ terms across 15+ languages for ID document field identification (NAME, DOB, EXPIRY, etc.)
- Document validation — confirms the captured image contains text consistent with a government-issued ID
What Runs through local AI models
The following processing happens in the user's browser and our servers during this analysis phase:
| Process | Details |
|---|---|
| Face detection + quality | Single face detection, centering, size, frontal pose validation |
| Liveness verification | Blink detection, head turn tracking, challenge completion |
| Anti-spoofing | Real score and live score computation |
| Document edge detection | Aspect ratio, coverage, and boundary analysis |
| Document OCR | Keyword extraction in 15+ languages via Tesseract WASM |
| Face descriptor extraction | 1024-dimensional numerical embedding from face mesh |
| Age estimation | Median age from 3+ face samples (age_only level) |
What Is Uploaded
After client-side processing is complete, the following artifacts are uploaded to AgeEvidence infrastructure for storage and admin review:
- Continuous video recording — video of the full verification session
- ID document frames — images of the ID front (and back, if applicable)
- Face descriptors — numerical embeddings (arrays of floating-point numbers)
- Challenge metadata — JSON data recording which liveness challenges were completed
- Fraud signals — anti-spoofing scores, EXIF data, and liveness scores
Does AgeEvidence Send Biometric Data to Third-Party Servers?
No.All biometric processing — face detection, liveness analysis, OCR, anti-spoofing — runs in the user's browser and on our servers using local AI models. Only the verification artifacts (video, document images, face descriptors) are uploaded to AgeEvidence infrastructure for storage and admin review. No biometric data is sent to third-party processors.
The entire analysis pipeline executes within the browser's JavaScript runtime using WebGL and WebAssembly. Network requests during the analysis phase are limited to loading the model weights (cached after first load). The upload phase begins only after all client-side checks pass.
Privacy Benefit
Client-side processing provides meaningful privacy advantages:
- Reduced data exposure — raw biometric frames (face images during analysis) are never transmitted over the network. Only processed artifacts are uploaded.
- No third-party dependencies — biometric analysis does not depend on external API services. There are no calls to cloud-based face recognition or OCR providers.
- User transparency — users can inspect network traffic to verify that no raw biometric data leaves their browser during the analysis phase.
- Regulatory alignment — processing biometric data client-side reduces the scope of server-side biometric data processing obligations under GDPR and similar frameworks.
Frequently Asked Questions
- Which browsers support client-side processing?
- Client-side models use WebGL and WebAssembly, which are supported in all modern browsers: Chrome, Firefox, Safari, and Edge on both desktop and mobile. Devices with a dedicated GPU will see better performance, but the models are optimized to run on mid-range mobile hardware as well.
- Does client-side processing affect verification accuracy?
- No. The same models and thresholds are used regardless of where processing happens. Client-side processing is a deployment choice that improves privacy without sacrificing accuracy. All quality thresholds (face size, centering, anti-spoofing scores) are enforced client-side before submission.
- Can I verify that biometric processing is happening client-side?
- Yes. The verification widget runs in an iframe that you can inspect with browser developer tools. Network traffic shows only the final artifacts being uploaded (video, image frames, numeric embeddings) — no raw biometric data is sent to any server during the analysis phase.