Aditya Kumar

Enterprise applications typically block sensitive topics like hate, violence or misinformation, but different departments may have unique policies — e.g., HR might allow finance queries, while IT doesn't. Without proper guardrails, AI applications risk responding to out-of-scope queries, leading to compliance issues. This session introduces a policy alignment framework for AI apps to enforce boundaries and prevent misuse. Learn how to implement app-specific policies seamlessly in Snowflake, using SPCS for containerized deployments, Model Registry for versioned policy enforcement, and compute pools for scaling and securing AI workloads.

RAG-based applications face AI-specific risks like prompt injections, PII leaks and toxic content. This session introduces the LLM Gateway, a framework that enforces guardrails, enhances trust and ensures compliance for safer AI interactions. Learn how the Snowflake data science team uses Snowflake Model Registry for seamless online and batch inference, and how Snowpark Container Services and GPU compute pools enable secure, scalable model deployment. Attendees will walk away with practical insights on registering, deploying and optimizing inference workflows for both real-time and large-scale generative AI applications — without external MLOps complexity.

Moving hundreds of thousands of players from on-premises data centers to the cloud with minimal downtime requires a unique mix of engineering, planning, and collaboration.