Kubernetes has become a foundation for modern cloud infrastructure. AI workloads are now pushing that foundation in new directions. The agenda covered a range of these challenges, from protecting KubeVirt VMs with Portworx Backup to understanding the infrastructure decisions behind AI workloads.
Seshu Kumar Mudiganti from Wind River looked at the terminology that Kubernetes engineers increasingly encounter, covering concepts such as training and inference, RAG, context windows, agents, quantization and LLMOps.
IBM’s Chinmaya Mishra and Pravin Ranjan explored another practical challenge: moving beyond conventional Kubernetes autoscaling to KV-cache-aware scaling for LLM inference.
A panel moderated by Sanil Kumar D brought together perspectives from Everpure, NeuralChaos.AI and VTAP AI Labs to discuss how AI is changing infrastructure and modern applications, including observability, security, cost and developer efficiency.
Together, these sessions highlighted that running AI in production is not simply a matter of adding GPUs to an existing platform. It introduces new requirements across the infrastructure and application stack.
SODA Foundation used the meetup to introduce its growing work around open-source AI infrastructure. The session “Energy and Context for LLMs: SODA Open Source AI Projects Unveiled!” brought together Sanil Kumar D from Caze Labs and Steven Tan from XFusion and the SODA Board/TOC.
The discussion introduced SODA’s open-source AI projects and looked at the importance of infrastructure context as AI workloads become part of production systems.
One of the projects highlighted was Aitra.
Aitra was featured in a dedicated session, “SODA AITRA Use Case and Production Deployment,” presented by Uday Swamy and Sandhya from Caze Labs/SODA. The session included a demonstration of AITRA Meter, its use case, and its integration with a real infrastructure-cost product.
Aitra is an open-source SODA Foundation project focused on AI inference energy efficiency and attribution. Aitra Meter provides Kubernetes-native measurement of GPU energy and connects infrastructure consumption with AI output, creating visibility at the workload, model and hardware level.
The project is being developed openly with the community, with contributions spanning measurement, energy and hardware providers, integrations, dashboards and documentation.
The Bengaluru meetup was also an opportunity to discuss where the community can contribute.
The AI infrastructure stack is developing quickly, and many of the problems are shared across organizations. Open source provides a way to work on those problems together, build common interfaces and tools, and learn from deployments in the real world.
For SODA Foundation, this is the purpose of bringing projects such as Aitra into the open: not simply to publish code, but to create a community around the infrastructure challenges that AI introduces.
The discussions in Bengaluru showed how broad that opportunity is, from Kubernetes and workload orchestration to inference, energy measurement, cost and infrastructure intelligence.
Thank you to Everpure for hosting the meetup, to all the speakers who shared their work and experience, and to the 90 attendees who joined the discussion.
We look forward to continuing these conversations with the Bengaluru community and building the next generation of open infrastructure together.
