Back to Speakers
Gaurav Srivastava - Director, Data Science Engineering at Adobe Inc

Connect

November 23–27, 2026

Gaurav Srivastava

Director, Data Science Engineering

Adobe Inc

Session Details

Behavioral Embedding Models to Combat Fraud/Policy Abuse and opportunities for monetization

My team has deployed and maintains several maintaining critical, production-level abuse detection systems that protect Adobe and its users. Some of those are: Piracy Detection: Broken File Signature, Checksum, Temper Guard Scraper Service: Hack Pattern Discovery, Reverse Engineering Account Sharing: Concurrent Usage, Impossible Travel Geo-price Arbitrage: Purchase vs. Usage Country Browser Fingerprinting: Attribute Hashing, Collision Resolution However, the threat landscape is evolving. Fraudsters are shifting from single, detectable events to complex attacks hidden in sequences of normal behaviour. This means our current detection, while effective, is fundamentally reactive. We are in a constant cycle of observing new, complex patterns and then manually building rules. This approach is slow to discover unknown threats at scale, limiting the inventory of abusers our team can act on. There is a need to build a proactive solution that continuously monitors the changing ground reality and detects emerging fraudulent behaviors of the users. Why embeddings? Create compact “behavior fingerprints” Learn multi-feature, temporal relationships Reusable universal features that can be easily used by downstream models Why transformers? Modelling on a sequence of events instead of aggregate feature values Bring self-attention and cross-attention capabilities needed to capture the relationships and temporal dependencies between events Works well even when labelled data is scarce (fraud/piracy space) Contrastive self-supervised learning technique to learn the embedding relationship allows the discovery of new fraud patterns and the detection of data drift

Full conference agenda

See every session across the AI Agents Conference November 2026 line-up.

View Agenda