Cloud Run is an ideal platform for hosting AI applications – for example, you can use Cloud Run with AI frameworks like LangChain or Firebase Genkit to orchestrate calls to AI models on Vertex AI, vector databases, and other APIs. In this session, we’ll dive deep into building AI agents on Cloud Run to solve complex tasks and explore several techniques, including tool calling, multi-agent systems, memory state management, and code execution. We’ll showcase interactive examples using popular frameworks.

NVIDIA GPUs accelerate batch ETL workloads at significant cost savings and performance. In this session, we will delve into optimizing Apache Spark on GCP Dataproc using the G2 accelerator-optimized series with L4 GPUs via RAPIDS Accelerator For Apache Spark, showcasing up to 14x speedups and 80% cost reductions for Spark applications. We will demonstrate this acceleration through a reference AI architecture on financial transaction fraud detection, and go through performance measurements.

Unstructured data makes up the majority of all new data; a trend that's been growing exponentially since 2018. At these volumes, vector embeddings require indexes to be trained so that nearest neighbors can be efficiently approximated, avoiding the need for exhaustive lookups. However, training these indexes puts intense demand on vector databases to maintain a high ingest throughput. In this session, we will explain how the NVIDIA cuVS library is turbo charging vector database ingest with GPUs, providing speedups from 5-20x and improving data readiness.

This Session is hosted by a Google Cloud Next Sponsor.
Visit your registration profile at g.co/cloudnext to opt out of sharing your contact information with the sponsor hosting this session.