At AMD, I architected PRISM, a multi-agent defect intelligence platform that connected Jira and Confluence for defect classification, report generation, and interactive Q&A. It automated gating and non-gating classification across ~5,000 Jira tickets per review cycle, replacing a manual weekly review.
I also took PRISM from local development to production, using a Snowflake RAG and text-to-SQL pipeline to ground answers in defect data and generate standardized executive reports. The platform used FastAPI, React, PostgreSQL, Terraform, and Docker across AWS and Azure.
At BookTranspo, I developed a multilingual customer-support chatbot using RAG and semantic retrieval, reducing manual support intervention by over 60%. I also built Python and SQL analytics pipelines and Power BI dashboards for route performance, vehicle utilization, customer behaviour, and revenue analysis.
