At Weedmaps, I’m leading a company-wide agentic analytics initiative that helps around 250 active users get answers to business and data questions. I built a custom MCP server over Snowflake semantic views and designed two evaluation tiers to test tool selection, routing, and refusals.
As the first data hire at Livepeer, I designed and built the data platform for a video infrastructure startup. A real-time Kafka and ClickHouse pipeline cut data latency from five minutes to 30 seconds, and the reliability feedback loop helped raise VOD reliability from about 65% to over 95%.
At Johnson & Johnson, I was the engineering lead on a data science team developing models to improve patient adherence across four pharmaceutical brands. I cut model RMSE by 23% and owned the team’s training and inference pipeline and internal Python package.
My work has also included predictive modeling for Walmart’s merchandising organization at Quantium and building the data science function from scratch at Black Hills Energy. Across these roles and projects, I’ve worked on forecasting, NLP classification, voter modeling, and retrieval systems, taking problems from data through evaluation and adoption.
