At Pratilipi, I redesigned Setu, an LLM pipeline that adapts stories across 12 Indian languages. I reduced defects by 58% using a seven-dimension quality benchmark I authored, and took an author-abandonment classifier from prototype to production on AWS.
At AI4Bharat (IIT Madras), I rebuilt multi-tenancy for 10 consumer apps and shipped a multi-model text-to-speech feature. My projects include evaluating a retail voice agent and building a local-first desktop dictation app.

