How YugabyteDB Empowers AI-powered Fraud Intelligence
summary
Claims a working pipeline that uses YugabyteDB with PGVector to store 384‑dimensional embeddings (HNSW index) alongside UPI transactions and run real‑time nearest‑neighbor search, calls an Anthropic model for semantic classification, and surfaces results via FastAPI/Streamlit. Demo handles queries like “donation > ₹10,000 in Chennai,” returns k nearest matches instantly, and records reviewer decisions into a feedback table so the model loop improves; system targets 18 billion monthly UPI transactions and demonstrates horizontal scaling and zero‑downtime replication.
AI-written summary · about
chapters (13)
- 0:00YugabyteDB and AI fraud intelligence
- 1:24UPI payment fraud risks
- 8:56Hybrid transactional-vector features
- 12:04Cloud-native to generative AI
- 15:15Vector embeddings in YugabyteDB
- 18:53Demo setup and schema
- 23:07Fraud classification UI
- 25:17High-value donation fraud
- 28:48Human feedback for fraud
- 31:17Odd-hour transaction analysis
- 33:43Travel and food fraud
- 35:30Model retraining with feedback
- 36:34Summary and benefits
related talks
26mEfforts Toward Adoption and Uncovering Hidden Features
33mThe Future of YugabyteDB : Vision, Innovation & Roadmap
25mRunning Shared AI Models with Avgidea Data Platform and YugabyteDB
23mUsing a Modern Database with ML to Prevent Financial Fraud
01mClip 5
22mBuilding a Modern Data Foundation: A Conversation with OCBC's Head of Data Architecture
14mLearn Vertex AI while building a fraud detection system
44mPostgres and OpenAI - a match made in Azure