At VesterAI, I was the founding AI hire and sole technical owner of the agent platform, taking it from zero to first paying customers. I reduced agent iterations from about 15 to 7, per-query cost by 89%, and latency by 37%.
I designed a CodeAct architecture and shipped a production memory system that carries user preferences, holdings, and prior insights across sessions. The agent achieved 78.5% accuracy on the Vals.ai finance benchmark, and I built checks for ground truth, claim verification, and answer quality.
Previously, at Amazon Robotics, I developed and deployed a PyTorch forecasting model that reached 91% average accuracy across production forecasts. At Dell Technologies, I built pricing and sales decision-support models; at Goldman Sachs, I built a data pipeline and dashboard workflow that eliminated roughly 20 hours of manual processing per week.
