At ZS, I co-designed and implemented a modular backend of five Azure AI Foundry-hosted agents, grounding workflows in Power BI Semantic Model data through FabricIQ knowledge retrieval. The system cut report turnaround from five days to two hours.
I also built an internal Agentic Copilot CLI that turns executive questions into Power BI dashboards, freeing 70% of analyst time. At ZS, I shipped a LightGBM sales propensity model that reduced CAC by 25%, and earlier improved engagement accuracy by 15% while cutting compute runtime by 50%; at IIT Bombay, I developed OCR for scanned Sanskrit documents that reduced Character Error Rate by 14%.

