Selected Work
I build AI/ML systems that make business data useful — from pricing and inventory decisions to large-scale risk prediction and document search. These three projects show how I approach the problem, develop a practical solution, and evaluate whether it works.
Featured Projects
Pricing & Optimization Systems
Decision science · Thrasio
Designed pricing and inventory decision systems that connected demand forecasting, price elasticity, and financial constraints. An inventory-aware pricing initiative increased profitability by 15% while reducing excess inventory by 20%. A separate price-elasticity workstream covered 10,000+ products.
Predicting Housing Complaints Before They Occur
Predictive modeling · Independent NYC HPD public-data project
Developed a quarterly complaint-risk model across buildings, apartments, and complaint categories, tackling rare events and limited prior history. In a retrospective evaluation of roughly 7 million previously generated scores, the highest-risk 10% showed 6.3× the overall complaint rate. This is measured ranking performance, not proven field-team impact.
RAG & Hybrid Search Product
AI search · Cadmus
Built and evaluated a hybrid search semantic document-retrieval foundation for a collection of approximately 10,000 documents. Compared three embedding models and selected MPNet based on retrieval relevance and operational tradeoffs. Hybrid search combines BM25 keyword retrieval with semantic vector search, then applies a reranker to refine the combined results and surface the most relevant documents.
Beyond the Featured Projects
I have also worked on production anomaly detection, drift monitoring and retraining workflows, large-scale data pipelines, and model calibration and evaluation. These capabilities support the end-to-end engineering required to put predictive and AI systems into practical use.
Have a Similar Problem?
Whether you need stronger forecasts, better business decisions, or searchable internal knowledge, I can help evaluate the opportunity and build a practical approach.