Business Analytics, Data Strategy, Operations
Business Analytics graduate (M.S., University of Miami) and Deloitte Scholar. I translate complex data into actionable insights through statistical analysis, predictive modeling, forecasting, and data visualization, working across SQL, Python, Tableau, and Power BI to drive measurable business outcomes.
Developing an AI solution for wildfire risk monitoring and early detection in collaboration with Deloitte, translating geospatial and environmental analytics into actionable insights for real-world decision-making.
My first published AI model: a LoRA fine-tune of Microsoft's Phi-3 Mini (4k) for chart reasoning, extracting structured insights from chart images. Trained with Unsloth + TRL and released on the Hugging Face Hub as a component of the SQL Agent stack.
A multi-model AI agent that accepts a natural-language question, generates and executes the corresponding SQL query, and returns an interactive data visualization. Three open-source LLMs were fine-tuned on a 761K-row curated dataset for SQL generation, chart reasoning, and SVG rendering, then deployed end-to-end on Hugging Face Spaces.
End-to-end big-data ML project on ~1.8M coffee-shop transactions. PySpark for ingest, cleaning, and EDA; scikit-learn + XGBoost + Optuna for modeling; SHAP for explainability. Delivered regression and classification models with clear lifts over baselines (wait time +34.8%, purchase amount +11.8%, rewards-target AUC 0.98).
Compared classical forecasting models, ARIMA, ARIMAX with exogenous regressors (layoffs, quits, sales, wages), Naive, and a Hybrid OLS+ARIMA, against a 12-month moving average baseline on monthly economic data. The simple moving average produced the lowest MAPE; a useful reminder that complexity isn't always better.
End-to-end binary classification project predicting customer behavior. EDA and preprocessing through gradient-boosted modeling with XGBoost, then SHAP values to explain individual predictions and overall feature importance. Final deliverable was a research poster.
Multi-month personal budgeting tool in Python with categorized expense tracking (needs vs wants), savings-goal progress, and data visualizations, pie chart by category, line chart over time, seasonal patterns, plus personalized tips against the 50/30/20 budget rule.