SNIA India T/E/N Storage Meetup - Bangalore - October - Hosted by HPE

Title: VFMOS: Virtualizing foundation models with a self-evolving operating system layer

Abstract: AI is moving from single foundation models (FMs) to compound agentic systems that combine multiple models, data sources, tools, and agents. Yet today’s agent frameworks/harnesses independently re-implement retrieval, model routing, context management, reasoning, and trust logic, creating a fragmented stack that lacks a principled way to jointly optimize intelligence and infrastructure resources, and share improvements across agents and workloads. As agents operate over longer horizons, this fragmentation increases cost and complexity and makes consistent governance harder.

In this talk, we argue that agentic AI needs a common systems layer analogous to the role operating systems and virtualization played for computing infrastructure: a Foundation Model Operating System (FMOS). FMOS virtualizes heterogeneous models, data, tools, reasoning, and verification behind stable Virtual Foundation Models (VFMs). Applications interact with a VFM through a standard model interface as if it were a dedicated, trustworthy foundation model, while FMOS transparently selects and composes the underlying resources and adapts how much context, reasoning, verification, and model capacity each request requires.

Unlike classical virtualization, which primarily preserves a stable abstraction over physical resources, VFMOS is designed for progressive quality and efficiency gains: learned improvements in knowledge, model selection, and trust can propagate across applications without requiring them to change. We will describe this OS-inspired architecture and discuss how virtualizing AI capabilities can enable more efficient, governable, and self-evolving agentic systems across heterogeneous AI infrastructure.

Followed by a SNIA Panel Discussion on Agentic AI in Enterprise Infrastructure: From Promise to Production — Challenges, Risks and Solutions
*Panel focus*: What is preventing Agentic AI from becoming reliable, secure, and operationally effective in real-world storage and data center environments?

Speaker:

Meetup Agenda:

2.00 to 2:30 pm Welcome /Registration

2:30 to 2:35 pm Welcome Address by HPE

2:35 to 2:45 pm SNIA Introduction by SNIA Board Member

2.45 to 3:45 pm VFMOS: Virtualizing foundation models with a self-evolving operating system layer by Tarun Kumar, HPE India Pvt Ltd

4:00 to 4:45 pm Panel Discussion on Agentic AI in Enterprise Infrastructure: From Promise to Production — Challenges, Risks and Solutions

4:45 to 5:00 pm Vote of Thanks

5:00 pm onwards Hi-Tea & Networking

Speakers:

Tarun Kumar, HPE India Pvt Ltd

Tarun Kumar is a Senior Research Engineer at HPE Labs, specializing in foundation model systems, agentic AI, and trustworthy AI infrastructure. He holds a PhD in Computer Science and Engineering from IIT Madras , where he received the Best PhD Thesis Award. He was also honoured with the Distinguished Alumni Award from Shiv Nadar University in 2026. Tarun is one of the lead proponents of the Virtual Foundation Model Operating System (VFMOS), an operatingsystem-inspired framework that introduces abstractions for managing foundation model-related processes. His work focuses on developing system-level foundations that enable more reliable, scalable, and trustworthy agentic AI systems and applications across enterprise environments.

SNIA Tech Speaker