Hi, I’m Shahbaz
I build LLM infrastructure and agent systems. The part I find interesting isn’t the model — it’s everything around it that has to work before an agent can do anything useful. Memory, state, config, boundaries, recovery, running things on the edge.
Before the current pivot, I spent time on quantum ML research and edge/embedded systems. The through-line is the same: I like the layer where messy physical or statistical reality has to meet clean software abstractions.
What I’m thinking about right now: runtime design for long-running agents, how to make memory actually useful instead of cosmetic, and pushing more of the agent stack onto local hardware.