Lead Data Scientist | GenAI Engineer

Designed and deployed an agentic AI prescription refill system orchestrating LLM reasoning, RAG-based context retrieval, PDF/eRx and doctor-note parsing, medical entity extraction, and clinical refill rules to automate medicine understanding, refill eligibility, and due-date calculation; scaled across 2.5M patients, improved operational throughput by 12%, generated 6% incremental revenue, and improved refill adherence by 3 percentage points. Built a LangChain-powered customer and competitor intelligence platform combining automated review/price scraping, sentiment modeling, and GPT-based theme extraction across 50K reviews; delivered CRM-ready competitor, pricing, and customer-experience insights that contributed to 20% NPS uplift, 12% repeat-rate improvement, and 8% incremental revenue. Developed a pharmacy churn prediction model using Random Forest, XGBoost, deep learning, RFM features, k-means segmentation, and survival/time-to-churn analysis; achieved 0.87 AUC and powered weekly CRM save programs across 1.5M customers.

Built time-series forecasting models (ARIMAX, SARIMAX, Prophet) for media impression and reach planning across audience segments; improved delivery against contracted reach targets by approximately 10% and reduced make-good obligations to clients. Developed a multi-class demographic profiling model using Word2Vec embeddings on content synopses and genres (XGBoost, RF); achieved 76% accuracy and enabled downstream semantic audience targeting. Built unsupervised audience segmentation (K-Means, DBSCAN) layered on profiling outputs; surfaced high-value clusters that improved campaign effectiveness by 20% and ROI by 15%. Migrated recurring analytical pipelines from PySpark to BigQuery SQL; cut pipeline runtime by approximately 50% and infrastructure cost by approximately 40%.

Designed and deployed a regression model to predict healthcare claim reimbursements with an R² of 0.89, streamlining financial forecasting for over 150K claims annually. Developed a high-performance classification model with precision: 92% and recall: 88% to identify billing inconsistencies, reducing manual review workload by 15%.