Somesh Zanwar · Data, analytics engineering, AI governance
Can this metric, workflow, or AI agent be trusted enough to make a decision?
I build the checks that answer that: data quality tests that become trust scores, policies that gate agents, and experiment readouts that say who to ship to. I finish an M.S. in Data Science at UT Arlington in December 2026.
Open to full-time · Dec 2026 · Arlington, TXEvery claim on this site is either linked to something public or marked as self-reported. This is the short version.
| Claim | Check it | Status |
|---|---|---|
| 5 merged pull requests in Microsoft's Agent Governance Toolkit | Public record | |
| An agent-control layer with signed single-use permits and 21 passing tests | Runs live | |
| An A/B readout where the aggregate says ship and two segments disagree | Runs live | |
| Improved reporting accuracy by 30% at IVS through validation and reconciliation | Self-reported |
I came into data work as an analyst. At IVS Software Solutions I reconciled operational records, defined KPIs with three teams and automated recurring reports. That is where I started caring less about the chart and more about whether the number under it could be defended.
Since then I have been building the checks that answer that question: data quality tests that become trust scores, trust scores that become incidents, and policies that decide whether a metric, or an AI agent acting on it, is allowed to proceed.
I am finishing an M.S. in Data Science at UT Arlington in December 2026, and I contribute to open-source agent governance work alongside it.