Aarohi Tiwari

At NVIDIA, I evaluated 1,000+ AI outputs across document, image, and video modalities, helping an AI evaluation and document intelligence programme maintain 97–98% evaluation accuracy.

I translate business needs into specifications, acceptance criteria, escalation triggers, and measurable publish/hold thresholds. My work cut rework by 70% and escalations by 30%, while helping a 70+ reviewer pipeline apply consistent standards.

I assess hallucination risk, define guardrails for edge cases, and turn defect trends into owned corrective actions through scorecards and Power BI dashboards. I also analysed 100+ financial reports to benchmark document intelligence models against regulated source material.

Previously at Vaibhav Global Limited, I reduced revision cycles by 60% by identifying recurring defects and fixing requirements at the source. I build practical evaluation systems that make AI quality, safety, and policy compliance measurable.