Bengaluru, Karnataka, India
I started in statistics — ISI, IIT Bombay, BHU — and over 6+ years it's grown into a career building production ML and analytics systems that drive real business decisions. Today, as a Senior Data Scientist at Eli Lilly, I work on Marketing Mix Models that quantify the impact of multi-million-dollar marketing spend on prescription volume across 210 U.S. DMAs. Beyond modeling, I solely designed and built Vertex — a full-stack Flask + AWS analytics platform now used by 150+ people across 4+ teams — and imBrain, a GPT-5 + RAG assistant that lets non-technical stakeholders query private reports in natural language. What I enjoy most: blending rigorous statistical methodology with the engineering work that turns models into things people actually use. Always interested in conversations around MMM, GenAI, and applied ML in pharma.
Impact Model — Marketing Mix Modeling | LightGBM, R, Python, Bootstrap, LOCO, missRanger • Contributed to R&D of an MMM framework using LightGBM to quantify DTC channel impact on NBRx across 210 U.S. DMAs for Lilly brands. • Engineered features from raw impressions, clicks, and spend — missRanger imputation, GRP normalization, and TV carryover lags — and validated via 10-fold geo-stratified cross-validation using mclust clusters. • Built the statistical inference layer with permutation-based p-values, bootstrap confidence intervals, and LOCO-based channel contribution, ROI, and mROI estimates. • Designed, developed, and deployed the end-to-end reporting pipeline and integrated it with the fully automated main modeling pipeline to generate standardized outputs each refresh cycle. • Owned the quarterly refresh — produced model reports, presented to U.S. brand and business stakeholders, and delivered recommendations that informed allocation of multi-million-dollar DTC marketing budgets. Vertex — Internal Analytics & Report Management Platform | Python, Flask, AWS S3, Docker, Quarto, Azure AD SSO • Solely architected, built, and deployed Vertex — a full-stack Flask/Python web application on AWS EC2 — now serving 150+ users across 4+ teams for all Lilly brands, centralizing report management and automated report generation. • Implemented role-based access control with Azure AD SSO via OAuth2 Proxy and Fernet encryption for sensitive files, metadata, and audit trails on S3. • Built Report Studio, a Quarto/RMarkdown execution engine packaged in a multi-stage Docker image that generates parameterized HTML/PDF reports per brand/quarter from YAML configs, cutting report turnaround from hours to minutes. • Designed a project-based collaboration model with three visibility tiers, in-browser file preview (CSV, HTML, Excel), and email subscription notifications — replacing fragmented email/SharePoint/shared-drive workflows. • Containerized and deployed the full stack via Docker
• Optimized multi-million-dollar marketing spend for healthcare, CPG, and retail clients by quantifying channel-level ROI and saturation across TV, YouTube, Facebook, Instagram, and other media using R and Excel. • Estimated channel contributions to annual revenue using multiple linear regression on adstock-transformed variables, and modeled ROI and spend-saturation curves via A/B/C parameters to inform optimal spend levels per channel. • Forecasted next-year revenue using time series models, helping clients plan annual marketing budget allocation; presented results via PowerPoint and Power BI dashboards. • Built dropout prediction models for U.S. educational institutions using XGBoost and ensemble techniques to estimate the likelihood of students leaving college, achieving high predictive accuracy. • Developed acceptance and grant allocation models predicting the probability that a student would accept an admission offer at a given grant amount; delivered weekly client reports based on model outputs.
• Contributed to an automated Value at Risk (VaR) calculation system for portfolio risk management using the Historical Simulation method. • Built interactive Power BI dashboards using DAX-based calculated columns and measures, and used advanced Excel (VLOOKUP, pivot tables) to present complex data clearly to stakeholders.