Autonomous fraud investigation driven by connected graphs.

Traditional AML models examine transactions in isolation. SentinelGraph navigates 4-hop money laundering paths, catches circular smurfing, and unmasks hidden infrastructure across FalkorDB in real time.

Autonomous Graph Investigation Console

FalkorDB Visual Topology
Account Frozen Person Company Device
Inbound AML Anomaly Dispatch
Agent Graph Tool Executions
Awaiting dispatch. Click button above to initiate multi-turn graph reasoning.
Explainable Audit Dossier
No investigation executed yet.

FalkorDB Property Graph Modeling

Nodes represent entities while edges represent verified financial, corporate, and digital interactions.

Account Nodes
Balances • Risk Scores • Frozen Status
Identity & Entity Links
Beneficial Ownership • Shell Corps • PEP Flags
Hardware Infrastructure
Device Fingerprints • Tor Exit IPs • Login Sessions

Graph Algorithms Exposed as Agent Tools

Parameterized Cypher queries executed directly against FalkorDB, equipping the agent with structural traversal, topological cycle detection, and live graph enforcement.

1. Multi-Hop Outbound Traversal
Fund Path
Victim → Mule Ring → Shell Co → Kingpin Sink
2. Circular Kickback Detection
Cycle Topology
Account A → Layering → Account A
3. Active FalkorDB Quarantine
Autonomous Action
Flagged Nodes → Mutate Status: FROZEN

Track 01: Agents That Act on Connected Data

SentinelGraph is purpose-built for the Graph Hacks Track 01 challenge. Rather than treating graphs as passive text embeddings, the autonomous agent uses FalkorDB's property graph as an active relational substrate for multi-hop reasoning, dynamic tool dispatch, and live database remediation.

Topology Over Isolation

Standalone transaction scoring misses smurfing loops and coordinated mule rings. By querying multi-hop relationship chains in FalkorDB, the agent perceives structural coordination across disparate entities.

Direct Graph Navigation

The LLM planner invokes specialized Cypher graph procedures—variable-length path traversal, cycle detection, and device/IP clustering—allowing the model to inspect live graph neighborhoods on demand.

Active Graph Remediation

When illicit rings are confirmed, the agent executes write queries against FalkorDB to freeze compromised accounts, isolate nodes, and record an immutable audit dossier with verified graph provenance.