01 Fragmented view
Each institution sees a different piece.
Local rule engines assess one transaction at a time. Coordinated activity can slip through because context stops at institutional boundaries.
Transaction Risk Intelligence
See the network.
Stop the fraud.
Detect coordinated fraud across banks, digital wallets, and payment networks—without centralizing raw customer data.
Graph-nativeSee relationships, not just transactions
Federated by designRaw data stays with each institution
Network Intelligence LIVE SIGNALS
Coordinated mule network4 cross-institution signals
94
“
Fraud moves as a network.
Risk engines often see only one institution.
RECOGNITION2nd PlaceBI–OJK Hackathon 2025
The blind spot
Fraud does not happen
in silos.
A single transaction can look normal. The pattern emerges when accounts, wallets, devices, merchants, and beneficiaries are read as one network.
02 Garda network view
Local signals form cross-ecosystem context.
Graph intelligence links entities and behaviors, revealing convergence, velocity, and layering earlier.
Interactive scenario
From normal transactions
to a network signal.
Run the analysis to see how cross-institution context changes a risk decision. The data below is a product simulation.
GARDA / INVESTIGATESIMULATIONLOCAL VIEW · 3 EVENTSNo coordinated risk detected
Ready to analyzeActivate federated graph context
!
94/100ANOMALY DETECTEDCoordinated mule network
Intelligence layer
One flow. More complete context.
Garda augments existing risk engines with relationships between signals that are usually hidden behind platform boundaries.
Detect
Identify transaction and behavioral anomalies within each institution.
Connect
Map relationships across accounts, wallets, devices, merchants, and beneficiaries.
Score
Score risk using patterns, proximity, velocity, and graph context.
Act
Send explainable signals to fraud teams and decision orchestration systems.
Privacy-preserving collaboration
Built for intelligence.
Designed for privacy.
Collaboration does not require a centralized customer data store. Garda is designed so models and risk signals move—not raw data.
- 01Local processing
Raw transaction data remains within each institution's environment.
- 02Federated intelligence
Model parameters and risk signals enable shared learning.
- 03Explainable outcomes
Every alert includes reasons a fraud investigator can review.
FEDERATED ARCHITECTURECONCEPT
A
Institution ACustomer data
▣B
Institution BCustomer data
▣C
Institution CCustomer data
▣Local Garda nodeLocal Garda nodeLocal Garda node
SHARED RISK SIGNALFederated
Intelligence
Intelligence
×
RAW CUSTOMER DATA DOES NOT MOVEPrivacy becomes an architectural property, not a disclaimer.
Designed for financial ecosystems
Find patterns no institution can see alone.
From transaction prevention to investigation, Garda helps risk teams see relationships and priorities faster.
PRIMARY USE CASE
Mule account network detection
Identify collection accounts, cross-bank convergence, layering, and linked cash-out paths.
Graph pattern Community & pathDecision Review / hold / alert
Scam network detection
Connect recurring accounts, phone numbers, devices, and beneficiaries across incidents.
Account takeover signals
Detect behavioral shifts and risky new relationships before funds move further.
AML investigation augmentation
Prioritize clusters, fund paths, and proximity to risky entities for investigation.
Transaction anomaly intelligence
Add graph context to existing rule engines and anomaly models.
Meaningful Intelligence
Meaningful intelligence should lead to better decisions.
Garda AI is the first product from Meaningful Intelligence—an Indonesian team building an intelligence layer for a safer, more collaborative, and privacy-conscious financial system.
View the pilot blueprint →2025RECOGNITION
02
2ND PLACEBI–OJK HackathonMeaningful Intelligence · Garda AI
Garda AI is an independent initiative by Meaningful Intelligence. References to BI, OJK, and BSPI 2030 provide ecosystem context and do not imply affiliation or endorsement.
From concept to controlled validation
Start with one fraud pattern.
Prove its value in a pilot.
Design a controlled pilot with synthetic or anonymized data, clear metrics, and privacy boundaries from day one.
01 Define pattern02 Connect signals03 Measure uplift04 Review governance