Jan 2024 – Mar 2025 · 1 yr 2 mos

• Designed and led the patient ranking system solution to suggest patients with undocumented conditions to clinical reviewers. Partnered with the clinical team for rapid feature engineering and deployed two step random forest models. The pilot generated savings of $4.2 million and paved the way for a wider implementation.
• Evaluated prompt engineering techniques on open-sourced healthcare large language models. This
• helped assess out of the box llm performance for missing patient diagnosis prediction task.
• Built DBT data pipeline, designed evaluation metrics and created a dashboard to integrate and evaluate an off the shelf AI product for ICD10 code diagnosis predictions. Complexity involved was in designing the pipeline to handle different comparisons involving common data rows that simplified and enabled fair comparisons with human generated diagnoses.
• Built high-cost risk stratification logistic regression model using claims and demographics data resulting in 15% improvement in recall over the rule-based methodology.
• Planned roadmap, did aggressive backlog pruning and simplified project success metric definition with clinical stakeholders to deliver quick alpha version in record time of two months.

Nov 2021 – Jan 2024 · 2 yrs 2 mos

• Built an entity recognition PoC pipeline with AWS Comprehend Medical api to tag diseases, symptoms, medications in physician notes with the end goal of using entities for downstream modeling tasks.
• Reduced production pipeline failures by 20% by using workarounds for data quality challenges by
• writing DBT data tests.
• Designed and implemented A/B test experiment to measure algorithm prioritized clinical review
• effectiveness over appointment/rule-based methodology. The 20% lift in new diagnoses in the highest
• category paved the way for release in all markets.

Aug 2020 – Oct 2021 · 1 yr 2 mos · Cupertino, California, United States

• Improved model prediction quality by 12% measured via event conversion by monitoring ML/AI models in production. Measured drift in training versus production data distributions by using metrics like Jensen Shannon divergence for categorical features and Kolmogorov-Smirnov test for numerical features.
• Improved end user adoption of predictions by 15% measured via increased user clicks by implementing model and individual prediction interpretability. Used Shapely values and clustering techniques to categorize features to facilitate interpretability. Leveraged this further to do weighted distance metrics that provided dynamic and interpretable similar entities to end users.
• Increased percentage of SaaS clients eligible for data science usecases by 12% by identifying common data patterns/issues across different clients using python and SQL. Generalized data cleaning & labeling of data across heterogeneous client datasets. This helped in smarter feature engineering for our models.

Jul 2019 – Aug 2020 · 1 yr 1 mo · San Francisco Bay Area

• ● Accomplished stability in model performance for small datasets by
• leveraging statistical techniques. The stability test helped qualify models
• that withstood small changes in data. This helped foster confidence in pilot
• users of the software.
• ● Assisted sales team by rapidly executing customer PoCs for various use
• cases such as churn reduction, lead ranking, free trial conversion prediction
• using multiple ML/AI models. Presented and interpreted model results to
• potential clients. Used learnings from PoCs to build functionality into a
• generic ml platform for domain users.
• ● Improved model performance by 5% by extensive error analysis to identify
• records where model was making mistakes. Iteratively improved
• performance by smarter feature engineering and better metrics for data
• quality.

May 2017 – May 2019 · 2 yrs

• Machine learning at Scale- Algorithm design in parallel computing (Using SPARK)
• Natural Language Processing with Deep Learning(Using Tensorflow)
• Experiments and Causality - Statistical Analysis with Experimentation and design based inference(Using R)
• Fundamentals of Data Engineering- Data Storage and Retrieval(SQLs, Tableau)
• Applied Machine Learning - Feature Engineering and Algorithms(Scikit learn libraries)
• Statistics for Data Science - descriptive and inferential statistics(Using R)
• Behind the Data:Human and Values(Legal, policy and Ethical implications of data)
• Research Design and Application
• Python for Data Science

Apr 2013 – Jan 2017 · 3 yrs 9 mos

• Conducted requirements analysis for reports and dashboards. Extensive client interaction in understanding user requirements; developed reports for user demo. Participated in defining dimensions and facts for star schema. Helped implement dashboards and migration to production.

Jan 2007 – Jan 2012 · 5 yrs

• Key contributor on client server development project with responsibilities to write detailed specifications, develop test plan and coding for individual modules.
• Designed for data model compatibility to support migration from legacy to new system by analyzing data schemas for both systems. Led a six-member onsite team for the design, development, testing and implementation of project SYNCH, a project designed to synchronize data from DB2 on IBM mainframe & MS Access to Oracle on UNIX.