• ▸Spotted that the existing credit scorecard was relying purely on static bureau data and missing early default signals. Trained XGBoost and logistic regression classifiers on 70K+ account records that now predict 90-day defaults 18% more accurately than the legacy model.
• ▸Engineered 45+ behavioral features from raw transaction history (repayment consistency, utilization trends, delinquency streaks) that give the model early warning signals the bureau score simply does not carry. AUC went up 11%.
• ▸Shipped a Streamlit explainability dashboard with SHAP breakdowns and segment-level lift charts that credit analysts and underwriters now rely on in every weekly portfolio risk review.
• ▸Currently re-architecting the risk summarization workflow into a multi-agent LangGraph system with dedicated agents for claims analysis, repayment behavior, and risk categorization, replacing a fragile single-pass RAG pipeline.
• ▸Containerized the full inference pipeline with Docker, deployed scheduled batch jobs on AWS EC2, and wired outputs through S3 into Tableau dashboards that leadership reviews every month.
• ▸Registered every model version and threshold decision in MLflow, validated them against FCRA and FDIC requirements alongside compliance teams, and built role-based access so analysts, underwriters, and reviewers each see exactly the data they need.

• ▸Enrollment teams were manually keying data off paper forms one field at a time. Built an OCR and SpaCy NER pipeline that automated the entire extraction and cut their manual review effort by 40%.
• ▸Wired ARIMA-based runtime forecasts directly into ServiceNow so the system auto-raised incident tickets the moment an ETL job looked likely to bottleneck. Saved support teams 6 to 8 hours of manual triage every single week.
• ▸Took full ownership of MLflow experiment tracking and model version control for the data science team, and handled all PHI data workflows in compliance with HIPAA.
• ▸Designed the team's core ETL infrastructure on AWS Glue and PySpark with Talend for orchestration, CloudWatch alerting, and CI/CD pipelines that supported both ML training and Tableau operational reporting.
• ▸Migrated legacy pipelines to Azure Data Factory with MuleSoft API integrations for EDI claims and enrollment flows, then built Grafana dashboards that let product owners monitor pipeline health themselves without ever pinging engineering.

• ▸Built SQL ETL processes to clean and consolidate de-identified patient records from multiple source systems, producing the datasets that fed clinical operations reporting reviewed by department leads each week.
• ▸Automated a full suite of recurring analytics reports in Python and SQL that the team had been rebuilding by hand every cycle. Cut turnaround time significantly and eliminated an entire class of manual copy-paste errors.
• ▸Delivered dashboards tracking patient readmission rates and cohort metrics that became the source of truth for clinical leadership in weekly operational reviews.
• ▸Partnered closely with senior data scientists on readmission risk models, handling data cleaning, de-duplication, feature validation, and every ad hoc data pull the modeling work demanded.

MS Data Science

University of Wisconsin, Milwaukee

BE Computer Science

Gandhi Institute of Technology and Management, Bangalore