Скачать или смотреть Designing Cost-Efficient Agent for Fraud Detection

הרצאה זו היא חלק ממיטאפ משותף עם חברת Cyera וקהילת MDLI
אתם יכולים לצפות בשאר ההרצאות ובמצגות פה: https://mdli.co.il/classifyai


In this talk, I will share insights from the system we built for Account Takeover (ATO) detection, which combines classification of events, texts, and documents for Risk Ops.


With an emphasis on the world of LLMs, the talk will focus on how to design a smart yet cost-effective system: when it is right to activate "heavy" models, when to settle for cheaper signals, and how to build decision layers that balance accuracy, latency, and cost.


I will present practical principles, including working with cascading models, RAG, prompt compression, and the integration of human analysts, in order to turn risk detection into a scalable, reliable, and economical system.