Enhancing Financial Analytics With Event-Driven Architectures and Graph Neural Networks
summary
Combines Neo4j with graph neural networks and an event-driven pipeline (Apache Kafka, ETL) to convert siloed SQL and unstructured data into live graphs and score transactions in real time. Claims a money‑laundering case that once required ~30,000 queries is detected by streaming graph scoring with example fraud scores of 75–83%, cutting false positives that cost banks ~$67M. Architecture enriches neighborhood features, issues instant alerts, scales on events for cost savings, and adds LLM narrations via LlamaIndex/LangChain for regulatory reports.
AI-written summary · about
chapters (8)
- 0:00Introduction to fraud analytics
- 1:12Impact and challenges of fraud
- 8:16Why real-time fraud detection matters
- 10:47Role of graph databases in fraud
- 15:38Event-driven architecture for real-time
- 19:34Graph neural networks for fraud detection
- 24:47Adding generative AI narrations
- 26:32Q&A and wrap-up
related talks
26mAdvancing Financial Fraud Detection with Graph Dbs
56mGraph Analytic Systems
25mReal-Time Fraud Detection at Scale
46mFraud Detection at Floa
12mCloud-Native Friendly Fraud Detection
13mBuild Real-Time Analytics Platforms for Financial Crime Detection
49mModern Graph Databases
18mTransforming Financial Decision Intelligence Through Advanced NNs