FINRA's knowledge graphs improve regulatory investigations
summary
FINRA builds a nightly pipeline that turns thousands of paper and digital records into graph triples using Amazon Textract, Amazon Comprehend and S3, bulk-loads them into Amazon Neptune and enriches with DMS/Kinesis and external web sources, then exposes navigable investigations with filters, timelines and source-highlighted evidence. Neptune ML and SageMaker run graph predictions; Neptune streams index results to Elasticsearch for full‑text search. The system reproduces known cases and uncovers additional links, cutting months of manual work.
AI-written summary · about
chapters (25)
- 0:00Intro
- 1:10Graphs are all around us
- 1:46When should I use a graph (database)?
- 2:21Thousands of customers are in production with knowledge graphs, fraud graphs, and identity graphs
- 3:08Amazon Neptune
- 4:11Neptune is globally available and secure by default
- 4:53Neptune: Key innovations
- 5:28Neptune ML
- 6:04Air routes dataset
- 6:36Neptune graph data model
- 7:19Amazon Kinesis integration
- 7:51AWS DMS support for Neptune
- 8:41How documents are processed today
- 10:28Amazon Comprehend
- 11:05Entity detection
- 11:36How FINRA visualizes documents
- 13:42Investigative work
- 14:21Investigation board
- 14:39The "John Doe" trail
- 16:06Investigative knowledge graph
- 19:57Architecture
- 23:24What we achieved
- 25:24Lessons learned
- 27:31Applicability of concepts to other regulation
- 28:30Next steps
related talks
32mFINRA's knowledge graphs improve regulatory investigations
15mGraphRAG-Powered AI Agents: Transforming Corporate Metadata into Intelligent Knowledge
28mFraud Graph with Amazon Neptune and Machine Learning
57mNeptune: Using graphs to gain security insights
27mEnhancing Financial Analytics With Event-Driven Architectures and Graph Neural Networks
30mModernizing data collection at FINRA with Amazon DocumentDB
26mAdvancing Financial Fraud Detection with Graph Dbs
56mGraph Analytic Systems