Sarosh Hussain

Python, TypeScript, Azure. Production AI systems, agent governance, CI/CD and evidence gating. Generalist by necessity, specialist in making agent output checkable.

Deterministic approaches to agent governance — what you can check without putting a model in the loop.

CrewScore — offline checker for missing written guardrails in agent system prompts. GitPin — PR gate that validates evidence claims at exact commit SHAs.

Talking about agent guardrails, prompt governance, and evidence-gated CI. Keen to hear about a CrewScore rule that fires when it shouldn't, or a control you'd expect that isn't in the 23.