Amritha Sharma

People Analytics & Hiring AI (PXT)

• Conducted extensive causal inference studies using TWFE, panel regression, and difference-in-differences models, evaluating the impact of internal AI tools, referral programs, and internship initiatives on hiring efficiency and success metrics.
• Developed a multi-agent AI framework using LangGraph, Strands, and Kiro, deploying agents for data extraction, causal analysis, document/narrative generation, and a Steering Guide Agent, integrated via AWS Lambda and connectors, automating insights for 200+ recruiters globally.
• Orchestrated scalable ML pipelines in LangGraph and Strands for automated reporting and advanced analytics, enhancing governance across global hiring workflows.

Advertising & Media Intelligence (Ads)

• Designed and deployed contextual bandits-based budget optimization models using Python/ML pipelines, dynamically reallocating ad spend across products and achieving ~40% lift in ROAS.
• Engineered a PySpark EMR pipeline tracking 17K advertisers’ behavior, enabling repeat usage framework that influenced >$2B in ad spend, with insights presented in executive WBRs.
• Developed reach estimation API for Prime Video and Twitch, utilizing probabilistic sketching and deduplication in scalable ML pipelines for cross-channel audience forecasting.

• Applied Bayesian inference and MCMC to analyze drivers of vaccine brand decisions for 5,000+ physicians, providing actionable insights that informed targeted campaigns and enhanced brand adoption.
• Identified 9M+ high-probability vaccination patients using ensemble methods and neural networks (recall 81%); applied tree-based segmentation on lifestyle data to optimize targeted marketing campaigns.
• Executed A/B tests for HPV and pneumococcal vaccine campaigns, testing targeting strategies and messaging, achieving 10× lift in campaign effectiveness.
• Developed model using peer-country similarity and engineered features to quantify vaccine hesitancy, facilitating identification of drivers and opportunities per country-level hesitancy percentage.
• Designed a survey automation tool using LLMs & sentence vectorization, enabling stakeholders to generate customer segments and marketing strategies within minutes.

• Implemented XGBoost classifier on 1.6 million personal care newsletters, identifying top 20 newsletters for R&D team based on predicted probabilities.
• Analyzed data from ~1000 sensor embedded devices using time series modelling to identify abnormal behavior in hair care product, providing recommendations to mitigate risk of failure.

• Managed daily operations of AWS enterprise data lake, assisting client in restructuring and optimizing data storage to achieve $3.5 million cost reductions.
• Managed daily operations of an AWS enterprise data lake and assisted the client in restructuring and optimizing data storage, resulting in $3.5 million cost reductions.
• Led team of 5 in ad hoc analyses for multiple breast cancer drugs and sales forecasting, achieving 12% reduction in MAPE.