Featured Project
Loan Performance Analytics & Risk Intelligence
End-to-end analytics platform built on 992K Freddie Mac loans ($248B portfolio) spanning 11 origination years. SQL-powered analytical engine with automated data quality checks, interactive Tableau dashboards, ML-based risk scoring, and auto-generated executive reports.
ML Risk Model — Results Explorer
Dual-model approach: origination predictor (honest AUC) + behavioral segmentation (risk ranking)
Top risk segments from the behavioral model. Segments with highest concentration of delinquent loans.
| Credit Band | LTV | Rate | Vintage | Loans | Risk Score | Actual DLQ |
|---|
The dataset is a single monthly snapshot, not a time series. True forward prediction requires sequential monthly data. The origination model predicts current delinquency status from borrower profile — valuable for risk stratification but not a deployment-ready forecasting tool.
Only ~1% of loans are delinquent. Precision is inherently low. Addressed with balanced class weights and evaluated using lift metrics rather than raw precision.
Monthly time-series data, additional features (employment, payment amounts, forbearance history), gradient boosting ensembles, and a calibration layer for dollar-loss estimation.
The first attempt produced AUC = 1.0 due to data leakage. The redesign separates Model B (origination, AUC ~0.77) for honest prediction from Model A (behavioral) for segmentation. This distinction is what mortgage analytics teams value.
Other Projects
Spotify Live Top 200 Dashboard
Power BI data intelligence platform — historical and real-time analytics engine built with API pipelines and Power BI modeling.
StatGuard: Security Patch Assistant
Can a static-only LLM-based agent safely assist with secure code review without ever executing the target program?
Tic-Tac-Tourney
Game logic with Random, Minimax, and Heuristic AI agents competing against humans in a round-robin tournament.
Psychometric Severity Prediction
Understanding how personality traits and demographics influence mental health outcomes using ML classification.
Evidence-Grounded RAG Chatbot
RAG pipeline using LlamaIndex and FAISS for document-grounded Q&A on IRS tax publications without external APIs.
Career & Education
Credentials & Badges
