GenAI Applications: Development to Production
Join us for an opportunity to network with AI applications developers in the Bay Area! Hear insights from speakers representing Nutanix, PingCAP, and others as they discuss the latest advancements in building and deploying scalable GenAI solutions.
Agenda
5:30 - 6:00 pm: Check-in/networking (food & drinks)
6:00 - 6:30 pm: Talk #1 - Lessons learned from managing GPU deployments on Kubernetes
6:30 - 7:00 pm: Talk #2 - Ship a Lightning-Fast FAQ & Stop Keyword Fails
7:00 - 7:30: Talk #3 - TBD
7:30 - 8:00: Wrap-up/networking
Talk #1: Lessons learned from managing GPU deployments on Kubernetes
Speakers: Sonali Mishra & Shalin Patel (Nutanix)
Abstract: Handling workloads that require GPU on Kubernetes has become easier with tools like NVIDIA’s GPU Operator but deploying them in production across real-world environments brings hidden challenges. In this session, we will share lessons learned from managing GPU across different infrastructures including on-prem and across public clouds, air-gapped environments, and different Operating Systems. We will cover compatibility issues with drivers and runtimes, discovering GPU attached nodes and scheduling GPU workloads on Kubernetes clusters. We will also discuss our experience working with vGPUs, the challenges of enabling multi-tenancy, dynamic resource allocation and monitoring. And lastly, we will talk about how we addressed some of these challenges using the NVIDIA GPU operator and our Kubernetes operator for vGPU tokens and license management. This talk is ideal for platform engineers and architects bringing AI/ML to Kubernetes, and looking to scale GPU use efficiently, securely and with better observability.
Talk #2 - Ship a Lightning-Fast FAQ & Stop Keyword Fails
Speaker: Chris Dabatos (PingCAP)
Abstract: In this session, learn how anyone can build & ship a lightning-fast FAQ with TiDB's built-in vector search with AWS Bedrock and a few plain English Python scripts. There's no need for developers to deal with microservices or cluster baby sitting, and since TiDB is MySQL compatible, your existing "ORM" will just work.