Piyush Agnihotri

At Accenture Strategy, I built and shipped an auditable compliance-scoring tool that checked roughly 600 regulatory obligations against bank documents, producing cited evidence and per-obligation scores. I also built complaint-triage and NLP taxonomy pipelines that turned large volumes of free-text complaints into actionable outputs for non-technical teams.

I've worked directly in the LLM evaluation loop at Turing, writing ground-truth reference solutions, rating model outputs, and diagnosing quantitative-reasoning failure modes. My applied statistics training at ISI Bangalore guides how I select metrics, validate outputs, and decide whether a model score can be trusted.