Deep Samal

Omnissa

Introspective Machines

Abstract

As large language models evolve into autonomous agents, understanding what they “think” becomes an engineering necessity. This talk surveys the interpretability stack — from mechanistic internals (sparse autoencoders, residual stream probes) to runtime guardrails — and extends the same defense-in-depth principles to multi-agent communication protocols.

Biography

I am a staff engineer in the performance engineering team at VMware/Broadcom/Omnissa. My team evaluates and optimizes performance of dynamic cloud-native Enterprise Mobility Management (EMM) applications used for managing and securing millions of Mobile, Desktop and IoT devices. I did my PhD under Saibal Mukhopadhyay and Marilyn Wolf at ECE from Georgia Tech. My research was focussed on Closed Loop Perception for Resource Efficient Autonomous System.