Kubeflow Commons Meetup #2
Join us at the Red Hat India office in Pune for an industry-focused AgenticOps meetup exploring the real-world engineering challenges of building and operating multi-agent AI systems at scale. Hosted by Kubeflow Commons Hub, this three-hour, practitioner-driven session digs into what it actually takes to move agentic workloads from experimentation into production.
Expect hands-on conversations about building evaluation pipelines with Kubeflow to automate reasoning-quality feedback loops, implementing real-time guardrails without constraining performance, and managing cost orchestration across complex agent workflows. We’ll also explore the MLOps realities of multi-agent systems from message-broker vs. shared-state communication on Kubernetes, to GPU contention across concurrent agents, to using distributed tracing to debug reasoning chains across microservices.
We’ll close with a discussion on AI sovereignty: why modern agentic architectures increasingly require air-gapped environments, and how Kubeflow and open-source tooling can help organizations get there with confidence.
Agenda
10:00 AM – 10:10 AM
Welcome & Opening Keynote
10:10 AM – 10:45 AM
Talk 1: Purnanand Kumar
From Hours to Minutes: Building Production-Ready Agentic Systems with IBM Fusion HCI, Red Hat OpenShift AI, and WatsonX Orchestrate
(30 min talk + 5 min Q&A)
10:45 AM – 11:20 AM
Talk 2: Shamsher Ansari
Operating Multi-Agent Systems Safely on Kubernetes using Agent Sandbox
(30 min talk + 5 min Q&A)
11:20 AM – 11:40 AM
Snacks & Networking
11:40 AM – 11:55 AM
Talk 3: Yash Agrawal
Running Megatron-Core LLM Training with Kubeflow Trainer
(10 min talk + 5 min Q&A)
11:55 AM – 12:30 PM
Talk 4: Vikram Shitole
Building Agentic AI Systems: Lessons from Real-World Deployments
(30 min talk + 5 min Q&A)
12:30 PM – 1:00 PM
Talk 5: Rankesh Kumar
Compliance Engine on Feast and MongoDB
(30 min talk + 5 min Q&A)
1:00 PM – 1:45 PM
Closing Remarks, Open Floor Networking & Speaker Interaction