At Infosys Limited, I architected a marketing analytics platform for TheHomeDepot that evolved from a monolithic LLM service into a hierarchical multi-agent GenAI system. It improved NL-to-SQL accuracy by 35%.
I also developed a schema-grounded hybrid RAG pipeline for the platform, reducing hallucinations by 80% and invalid SQL by 30%. Centralized state management and context artifact caching cut redundant LLM calls, cost, and latency by 30%.
For TheHomeDepot, I worked on an AI-powered Slack support bot that reduced manual support effort by 50%. Fine-tuning GPT-4 Turbo on 6,500 conversations supported a human-in-the-loop review workflow that reached 65% adoption and reduced resolution time from 15 minutes to 2.
Earlier, as a Machine Learning Intern at Bilateral Solutions, I built voice-based conversational AI systems for enterprise clients, including multilingual ASR/LLM voicebots for Domino’s hiring. I also engineered and operated Apigee cloud API platforms, and my projects include a multi-document Q&A RAG pipeline and a CNN image-classification model.
