At Egnyte, I shipped an LLM routing system that cut inference costs by 43% across 650K+ monthly requests and re-architected a spreadsheet Q&A ReAct agent into SQL generation, reducing cost and latency 10x. I also led a GCP Vertex AI multi-label AEC tagger to 95% weighted F1 and improved search quality by 31%.
Previously at Nanos.ai, I fine-tuned LLMs for Google Ads keyword generation, built RAG and multi-agent systems that reduced manual overhead by 40%, and applied bandit optimization to raise campaign efficiency by 20%. My work spans LLM evaluation, fine-tuning, agent orchestration, ML modeling, computer vision, and production deployment.
