Selected work
The strongest applied AI work often happens inside private operating environments, so this page combines representative enterprise outcomes with public code that shows how I think and build.
Redesigned large-scale forecasting across a fulfillment network where locations, products, promotions, and inventory patterns changed constantly.
Improved site-level forecast error from 30%+ to under 15% while supporting downstream operational planning.
Built and improved pricing systems used in live shipment workflows where latency, uptime, and interpretability all mattered.
Matched best-ensemble performance with a compact model design based on five common shipment inputs.
Improved recommendation workflows and paired them with explanations that gave end users more confidence in model-driven suggestions.
Helped turn model output into usable sales and operational decisions instead of opaque predictions.
A Streamlit application that consumes and caches live market data, visualizes historical price movement, and compares technical and statistical forecasts.
A compact example of end-to-end product thinking across data ingestion, modeling, and UX.
Recommendation engine work built from the Goodreads UCSD dataset, exploring practical ranking and user preference signals.
A simple public artifact that mirrors my longer-running interest in recommendation and decision support systems.
This site itself: a lightweight Flask application that presents consulting positioning, structured SEO metadata, and a modern profile for AI/ML work.
Designed to communicate clearly with both human buyers and search crawlers while staying easy to maintain.