Projects
A collection of my work and side projects
Featured Projects
Agentic NL2SQL System
Engineered multi-agent architecture that achieves 50.87% execution accuracy on BIRD benchmark while reducing inference cost by 90%. Preserves privacy using locally finetuned SLMs through LoRA & QLoRA that can run on consumer-grade hardware. Accepted at CAI 2026 Conference.
Interpretable Credit Risk Models
Built interpretable credit risk models using Ensemble Models + LIME that achieve 76% accuracy while ensuring regulatory compliance and full model transparency. Published at CADSCOM 2024 & Accepted at AMLDS 2025.
Insurance Predictive Modeling
Developed rare-event classifiers for Travelers Insurance call center staffing needs using zero-inflated regression and advanced feature engineering on highly imbalanced data. Evaluated with Gini index for production model selection. EDA revealed Poisson-distributed call data, motivating zero-inflated regression to predict policyholder demand and prevent call center over/under-staffing.
Other Projects
2022 World Cup Performance Analytics
Scraped and processed FIFA statistics; built dashboards with radar plots and bar charts analyzing team performance metrics for the 2022 World Cup.
REAL TIME CUSTOMER CHURN PREDICTION
Built a real-time churn prediction platform: trained a Random Forest pipeline, streamed events via Kafka, processed features in Spark, and served predictions through a React dashboard. Why it is Important Customer acquisition costs 5-25x more than retention. By identifying at-risk customers early, businesses can deploy targeted interventions like personalized offers, proactive support, or contract adjustments to reduce churn and protect revenue.