I'm a Data Scientist with a Master's in Data Science from Merrimack College specializing in end-to-end ML pipelines, agentic AI systems, RAG applications, and large-scale data analysis. My projects work with real-world datasets — 16M+ row U.S. Census data, 4.25M HMDA mortgage applications, and 941K NYC 311 service requests. I build in Python, SQL, and R, orchestrate pipelines with Apache Airflow, deploy LLM-powered applications with the Anthropic Claude API, and hold an AWS Cloud Practitioner certification. Background in legal operations and data governance. Currently seeking entry-level Data Scientist roles.
- 🛒 ReviewPulse — Agentic AI Consumer Insight System — Autonomous LLM agent (Claude tool use) that investigates product review data, detects statistically significant defect trends via spike detection, and generates evidence-backed action reports; includes a Streamlit dashboard with an interactive Q&A assistant grounded in the same agent tools
- ⚖️ CFPB Legal Q&A RAG Agent — Retrieval-Augmented Generation pipeline built with LangChain, ChromaDB, and the Anthropic Claude API; ingests the CFPB Supervision and Examination Manual, retrieves semantically relevant passages via cosine similarity search, and generates cited, grounded answers to mortgage compliance questions via a Streamlit chat interface
- 🏦 Home Loan Approval Prediction — ML pipeline on 4.25M real HMDA 2023 mortgage applications; XGBoost ROC-AUC 0.9932, 96.3% accuracy across 121 features; SHAP-based feature importance and fairness analysis across demographic groups
- 🏠 Rent Burden Prediction — Fairness & ML analysis on 16M+ ACS PUMS household records (Logistic Regression, Random Forest, Gradient Boosting); equity analysis across race, sex, and geography for HUD policy context
- 🔄 HMDA Mortgage Data Pipeline — Apache Airflow — Orchestrated weekly ETL pipeline using Apache Airflow (Astro CLI) on 279K NC HMDA 2023 mortgage records; automated ingestion, cleaning, quality validation (4 data checks), and model performance monitoring as a DAG; AUC 0.7273 on held-out test set