At Synopsys, I built a multi-agent system that diagnoses chip-design issues 5x faster than the manual process it replaced, matching expert-identified root causes on the cases evaluated so far. I also developed real-time detection for stalled design runs while keeping overhead under 1% on healthy runs.
I built an MCP server that answers questions about VLSI design data in plain English and paired semantic search with an open-source LLM for context-aware answers. My physical-design root-cause pipeline uses a property-graph knowledge graph to help AI agents identify causes and generate ready-to-run fix scripts.
Earlier, at Synopsys, I contributed core parts of DesignDash and developed models for chip-design decisions, including cell drive strength and threshold-voltage prediction. At Novartis, I built a production forecasting pipeline projected to have a $20M impact and created an NLP pipeline to extract protein and target information from clinical-trial documents.
