Uploaded August 2026 | Updated September 2026, 2 weeks ago
How do you detect complex fraud patterns when transaction data is spread across disconnected systems?
In this session, Craig Shallahammer, Oracle ACE Director, builds a fraud-detection pipeline using Oracle AI Database, SQL, PL/SQL, vector search, embeddings, and property graphs.
The example follows “Slick Eddie,” a fraudster using gift cards across three disconnected transactional systems. The challenge is to identify the fraud pattern without relying on foreign keys or processing the data with Python.
The session demonstrates how to:
- Create a fraud-fingerprint embedding
- Use vector similarity search to identify suspicious transaction activity
- Analyse connected transaction paths with property graphs and PGQL
- Reduce millions of possible paths to a manageable set of candidates
- Apply fuzzy balance and timing checks to validate the fraud sequence
- Generate actionable fraud alerts from the resulting evidence
The final workflow reduces 3.7 million possible paths to 132 candidates and identifies the transaction sequence that matches the fraud pattern.
This is not a polished “happy path” demo. It is a practical account of what worked, what failed, and how an AI coding assistant helped build a multi-phase fraud-detection workflow in days.
Oracle AI Database brings SQL, vectors, and graphs together in one engine, helping developers build more complete analytical and AI applications without moving data across multiple systems.
Learn more about Oracle AI Database:
oracle.com/database/ai-database
Explore Oracle Graph:
oracle.com/database/graph
AskTOM:
asktom.oracle.com
Chapters:
00:00 Welcome and session introduction
04:40 The Slick Eddie fraud scenario
11:45 Why the fraud pattern is difficult to detect
27:00 Designing the fraud-detection workflow
32:00 Property graphs, PGQL, and transaction-path analysis
42:55 Vector similarity search and fraud fingerprints
55:00 Results, validation, and lessons learned
58:25 What failed and what the AI coding assistant changed
01:02:00 Q&A and discussion
#OracleAIDatabase #VectorSearch #GraphAnalytics #FraudDetection
How do you detect complex fraud patterns when transaction data is spread across disconnected systems?
In this session, Craig Shallahammer, Oracle ACE Director, builds a fraud-detection pipeline using Oracle AI Database, SQL, PL/SQL, vector search, embeddings, and property graphs.
The example follows “Slick Eddie,” a fraudster using gift cards across three disconnected transactional systems. The challenge is to identify the fraud pattern without relying on foreign keys or processing the data with Python.
The session demonstrates how to:
- Create a fraud-fingerprint embedding
- Use vector similarity search to identify suspicious transaction activity
- Analyse connected transaction paths with property graphs and PGQL
- Reduce millions of possible paths to a manageable set of candidates
- Apply fuzzy balance and timing checks to validate the fraud sequence
- Generate actionable fraud alerts from the resulting evidence
The final workflow reduces 3.7 million possible paths to 132 candidates and identifies the transaction sequence that matches the fraud pattern.
This is not a polished “happy path” demo. It is a practical account of what worked, what failed, and how an AI coding assistant helped build a multi-phase fraud-detection workflow in days.
Oracle AI Database brings SQL, vectors, and graphs together in one engine, helping developers build more complete analytical and AI applications without moving data across multiple systems.
Learn more about Oracle AI Database:
oracle.com/database/ai-database
Explore Oracle Graph:
oracle.com/database/graph
AskTOM:
asktom.oracle.com
Chapters:
00:00 Welcome and session introduction
04:40 The Slick Eddie fraud scenario
11:45 Why the fraud pattern is difficult to detect
27:00 Designing the fraud-detection workflow
32:00 Property graphs, PGQL, and transaction-path analysis
42:55 Vector similarity search and fraud fingerprints
55:00 Results, validation, and lessons learned
58:25 What failed and what the AI coding assistant changed
01:02:00 Q&A and discussion
#OracleAIDatabase #VectorSearch #GraphAnalytics #FraudDetection
