📚 AI Training Data

Click any scenario to visualize 500 training samples on 3D space

💡 Click any 3D point to inspect its full 40D IR × 40D CR vectors

🚀 Live Simulation

1 Million transactions over 1-hour window

⚙️ Configuration

Realistic Distribution (Industry Standard)
Clean ~94% Review ~5% Fraud ~1%
⏱️ Fraud rates vary by time: higher at night, weekends

📊 Live Metrics

0
Processed
0
Tx/Second
--
AI Accuracy
0ms
Avg Latency
⚠️ AI not trained - Click "Train AI First"
3D Risk Space Visualization
Transactions mapped from 80D to 3D • Perfect Vector at origin
Points displayed: 0
Simulated Date/Time
-- --:--:--
1-Hour Window Progress
0m 0s / 60m
ACCEPT (Low Risk)
REVIEW (Medium Risk)
BLOCK (High Risk)
Perfect Vector (Origin)
🖱️ Click any transaction to inspect 40D vectors
📐 Decision Boundaries
Accept Zone
r = 3.00
—
Review Zone
r = 6.00
—
Block Zone
r = 9.00
—
Boundaries adapt with incoming data
Awaiting transactions

🤖 Multi-Agent Ensemble

Ready to simulate 0%
0
Accept
0
Review
0
Block
1. Similarity Agent 40%
Cosine similarity vs Adaptive Perfect Vector. Adjusts Accept/Review/Block boundaries based on deviation from ideal behavior.
--
2. Anomaly Agent 🟣 25%
Mahalanobis + Isolation + LOF ensemble. Flags outliers independently of boundaries, even in green zone (purple dots).
--
3. Neural Agent 🩷 35%
80→64→32→1 network with online learning. Detects non-linear patterns and boundary-region outliers (pink dots).
--
4. Confidence Agent Meta
Variance-based agreement. High disagreement → REVIEW. Prevents false confidence.
--
5. Explainer Agent Output
Top risk factors with contribution scores. GDPR Art. 22 compliant.
Awaiting analysis...

Agent Details