Problem + User
- Fraud operations teams need graph-level context, not only isolated transaction scores.
- Analysts need explainable reasons and fast drill-down for alert investigation.
Graph-native Fraud Monitoring
Graph Fraud Command Center
Real-time graph-risk monitoring with simulation controls, explainable alert reasons, and API-level observability. This raw HTML includes a seeded operational snapshot for non-JavaScript clients and is hydrated with live data when JavaScript is available.
Architecture
- Transaction events -> graph feature extraction -> risk scoring API.
- Alert path -> reason attribution -> investigation table and runtime panel.
- Metrics pipeline -> Prometheus -> dashboard refresh and monitoring views.
SLIs / Signals
- Score request throughput and high-risk ratio.
- Mean HTTP latency and dashboard refresh latency.
- Runtime health status and latest alert distribution by band.
API reachable (snapshot)
Ready for simulation. Use "Run Simulation" to generate additional events.
Operational Snapshot
Graph Nodes
28,941
Unique accounts
Graph Edges
186,402
Observed relationships
Events Total
942,188
Transactions processed
Alerts Total
3,682
Flagged transactions
High Risk (1h)
27
Score >= 0.85
Score Requests
18,540
Prometheus counter
High-Risk Ratio
2.33%
high_risk / score_requests
Mean HTTP Latency
19.40 ms
From histogram sum/count
Latest Alerts
| Time | Tx ID | Route | Amount | Score | Band | Reasons |
|---|---|---|---|---|---|---|
| 3/1/2026, 4:51:29 PM | tx_9fb1c2e4 | acct_0142 -> acct_8821 | $14,920.55 | 0.941 | critical | • velocity_spike • high-risk corridor • beneficiary novelty |
| 3/1/2026, 4:50:11 PM | tx_3ad59b77 | acct_7751 -> acct_1204 | $8,260.00 | 0.887 | high | • structuring pattern • device risk shift |