VEIL-X

v1.0
Visual Edge Intelligence Layer
ACTIVE
Connecting...
Execution Pipeline Ready
○ Tab Capture
○ DOM PII Detect
○ Visual Privacy Scan
○ UI Perception
○ Region Fusion
○ Client Redaction
○ Local Vision ONNX
○ Server Reasoning
○ Browser Action
PRIVACY GATE & DIAGNOSTICS GATE: IDLE
DOM detections:0
Visual detections:0
UI elements:0
Fused sensitive regions:0
Regions redacted:0
Visual scan status:READY
Pixel verification:--
Raw image transmitted:NO (Client Memory Only)
Email: 0 Phone: 0 Password: 0 PIN: 0 Aadhaar: 0 Card: 0 Face/Photo: 0

Latency Breakdown (Total E2E: 0ms)

Tab Capture:
0ms
DOM PII Detection:
0ms
Visual Privacy Scan:
0ms
Region Fusion:
0ms
Canvas Redaction:
0ms
Local ONNX Vision:
0ms
Server VLM & RTT:
0ms

Local Visual Privacy Engine (On-Device Perception)

Visual Privacy ACTIVE
Engine Pipeline Tesseract WASM + UltraFace ONNX
Engine Footprint 3.25 MB
OCR Latency 0ms
Face Model Time 0ms
Fusion Time 0ms

Client Compute & Memory

ONNX Backend WASM (SIMD)
Peak JS Heap Memory -- MB
System CPU (run avg) --
Scene Classifier SqueezeNet (4.95 MB)
Input Tensor 1 × 3 × 224 × 224

Local On-Device Vision Output

Genuinely computed in browser via onnxruntime-web without server calls.

Run an agent action to see real-time local vision predictions.

PII Precision & Recall Benchmark

Evaluates DOM, Visual OCR, and Fused detection against ground-truth sensitive regions.

Detection Precision --% TP / (TP + FP)
Detection Recall --% TP / (TP + FN)
F1 Score --% Harmonic Mean
Redaction Precision --% Sensitive Area / Redacted Area
Redaction Recall --% Covered Area / Target Area
Fusion Verification DOM + Visual Dual-Layer Protection

Confusion Matrix Metrics

True Positives (TP): --
False Positives (FP): --
False Negatives (FN): --
True Negatives (TN): --
Target / Item Source Layer Sensitive Detected Verdict
Click "Run Benchmark" while viewing the demo portal or document viewer.