At Three Verticals, I built multi-agent AI systems that automated 12 internal workflows, reducing average task time from 4.2 minutes to 55 seconds. I also designed a Pinecone search system with hybrid retrieval and re-ranking that raised answer accuracy from 71% to 94% across 4.8 million documents.
I created model routing and caching across LLM providers, reducing monthly AI API costs by 41%. I optimized serving with vLLM and quantization to deliver 87ms response time at a peak load of 1,200 requests per second.
Earlier, at Inceed and TEKSystem, I built production ML models, training systems, and data pipelines. My work included reducing false alarms by 34% and shortening model training time from 38 hours to 6.5 hours.
