At Univo Education, I implemented a Gemini 2.5 Flash pipeline that scored more than 12,000 sales calls against a 78-field behavioral schema. I also trained conversion models and analyzed objections, finding that direct responses resolved them more often than deflection.
At Indian Political Action Committee (IPAC), I fine-tuned XLM-RoBERTa for multilingual sentiment classification, reaching 92% weighted F1 across four languages. I also engineered a LangGraph and GPT-4 document pipeline that processed more than 80,000 PDFs at 95%+ accuracy, eliminating 20 FTEs of manual data entry.
I designed batch orchestration that cut processing time by 80% and shipped a personalization pipeline delivering more than 45,000 videos. My work also includes audio fraud detection and analytics for voter verification and health campaigns.
