About
I'm a MS Business Analytics candidate at the University of Washington Foster School of Business (GPA: 3.85, Top 10%) with 3+ years of experience transforming data into business strategy. I've built ML models that identified 847 vulnerabilities across 50K+ endpoints, generated $672K in incremental revenue through territory optimization, and automated reporting pipelines that save 1,200+ work hours annually. My work spans predictive modeling, causal inference, and business intelligence — always focused on the question that matters most: what should we do next?
Skills
Experience
Featured Projects
Real-world analytics projects solving business problems — from causal inference and A/B testing to ML models and LLM applications.
Business Problem: Do display ads actually drive conversions across 576K users? Designed a randomized controlled trial with ITT/ATT analysis and segmentation by user type and geography. Impact: Identified a 2.2% conversion lift in inactive users (89% of the base) and a 3.67% lift in non-American users — recommending targeted spend reallocation to improve ROI.
Business Problem: Was Google Maps advertising actually driving incremental store visits across 200 US markets? Applied Difference-in-Differences and Synthetic Control methods across 200 DMAs, controlling for market characteristics. Impact: Proved naive last-click attribution dramatically overstated ROI — enabling accurate multi-million dollar budget reallocation.
Business Problem: Which ad impressions are actually worth buying? Built CART and XGBoost models on 80K+ impression-level observations to forecast click-through rate. Impact: Recommended targeting the top 5K impressions for a 5.5x ROI versus the untargeted baseline.
Business Problem: How do you automate routing and escalation of customer service tickets at scale? Fine-tuned GPT-4.1-mini on 67 curated examples to generate structured JSON outputs with issue tags and safety flags. Impact: Significantly improved classification accuracy and escalation detection versus prompt engineering alone.
Business Problem: How do consulting teams find relevant past solutions without reading thousands of documents? Built an end-to-end RAG system using GPT-4 to semantically search 3 years of project documentation. Impact: 65% resolution rate on standard queries, sub-3-minute response time, reduced hallucinations versus plain LLM baseline.
Business Problem: Can we trust observational data to make attribution decisions worth millions? Compared IPW, Double ML, and adjusted regression against experimental ground truth on a 2M-user advertising dataset. Impact: Double ML achieved 83% accuracy while naive attribution overestimated lift by 148% — a critical finding for ad spend decisions.