Research Paper
LLMs + Knowledge Graphs: Building the Cognitive Architecture for Human-Like AI
Coming Soon
Project Overview
Hybrid Knowledge Graph-Powered Fraud Detection
Sentry (Suspicious Entity Network Tracking & Analysis) is a hybrid knowledge graph-powered fraud detection system designed to identify complex, multi-entity fraud schemes that traditional SQL-based systems often miss. By modeling relationships between accounts, devices, IPs, and merchants, Sentry enables multi-hop reasoning, allowing analysts to uncover hidden fraud networks, such as rings of accounts transacting with multiple flagged entities.
The system integrates PostgreSQL for high-throughput transactional data and Neo4j for relationship reasoning, orchestrated through LangChain and OpenAI for natural language to Cypher query translation. Its natural language interface lets users ask plain-English questions and receive explainable, graph-based responses complete with audit trails for compliance. Built with modularity and enterprise-level scaling in mind, Sentry is deployable on AWS and positioned as a production-ready prototype for fraud teams looking to move beyond brittle rule-based detection toward trustworthy, AI-powered analysis.
Sentry Dashboard
Placeholder for main dashboard screenshot
showing fraud detection interface and graph visualization
Key Features
Multi-Hop Relationship Reasoning
Uncovers hidden fraud networks through complex entity relationships, rings of accounts transacting with multiple flagged entities
Natural Language Query Interface
Ask plain-English questions like 'Which accounts have used flagged IPs in the last 30 days?' and get graph-based responses
Hybrid Database Architecture
PostgreSQL for high-throughput transactions + Neo4j for relationship reasoning, orchestrated via LangChain and OpenAI
Explainable AI & Audit Trails
Bridges semantic gap between business questions and underlying schemas, ensuring deterministic, auditable outputs for compliance
Live Demo
Experience Sentry
Get hands-on experience with Sentry's hybrid knowledge graph fraud detection system. The demo showcases how users can ask plain-English questions and receive explainable, graph-based responses that reveal hidden fraud networks.
Live Demo Video
Placeholder for Sentry demo video
showing AI-powered fraud detection and graph analytics
Architecture & Screenshots
Hybrid Graph Architecture
Hybrid Graph Architecture
PostgreSQL for transactions + Neo4j for relationships
orchestrated via LangChain and OpenAI
Fraud Detection Workflow
Multi-Hop Fraud Detection
Uncovering hidden networks through
complex entity relationships
Knowledge Graph Schema
Entity Relationships
Accounts, devices, IPs, merchants
Query Interface
Natural Language Queries
Plain-English to Cypher translation
Audit Trail System
Compliance Reports
Placeholder for audit trail generation
Technology Stack
