Aman Kushwaha Email and Phone Number

Dynamic Targeting and Portfolio Optimization• Developed and deployed an algorithm that dynamically recommend which physicians to target based on their behavior and maximize the overall sales• Built multiple non-linear tree based boosting model that measure brand-wise opportunity such as physician writing behavior. Change in behavior and look-alike writers to increase individual drug adoption• Designed an optimization on top of this (using PuLP) to recommend brand wise details by… Show more Dynamic Targeting and Portfolio Optimization• Developed and deployed an algorithm that dynamically recommend which physicians to target based on their behavior and maximize the overall sales• Built multiple non-linear tree based boosting model that measure brand-wise opportunity such as physician writing behavior. Change in behavior and look-alike writers to increase individual drug adoption• Designed an optimization on top of this (using PuLP) to recommend brand wise details by avoidingcannibalization and maximizing the salesEntity Extraction with LLM and RAG• Constructed an end-to-end RAG pipeline utilizing LLM (GPT-4) for entity extraction• Implemented Azure Form Recognizer and pdfplumber for efficient document digitization and OCR• Employed LangChain for document chunking, prompt template creation and NLP tasks• Utilized prompting techniques like Few-Shot and Chain of Verification, achieving an F1 score of 0.90Protocol Authoring Tool• Built a web application using dash to streamline the authoring of protocol documents for clinical trials• Incorporated Large Language Models (LLM) to provide high-quality recommendations based on user inputs and instructions, significantly minimizing manual effort by 30% in drafting the first draft Show less

Embedding Generation for Emerging Pharma• Engineered a feature creation pipeline capable of generating 1.5k-2k features, utilizing feature engineering techniques and data preprocessing• Implemented Autoencoder to generate lower-dimensional embedding. Employed Optuna to optimize critical parameters, including the number of layers, neurons per layer, and embedding size.• Applied the derived embeddings to various HCP level use cases, achieving an impressive 20% lift in outcomes… Show more Embedding Generation for Emerging Pharma• Engineered a feature creation pipeline capable of generating 1.5k-2k features, utilizing feature engineering techniques and data preprocessing• Implemented Autoencoder to generate lower-dimensional embedding. Employed Optuna to optimize critical parameters, including the number of layers, neurons per layer, and embedding size.• Applied the derived embeddings to various HCP level use cases, achieving an impressive 20% lift in outcomes compared to conventional approachesImpact Modeling Pipeline using Causal Inference• Developed an impact modeling pipeline leveraging causal inference methods (S-learner and T-learner) to quantify the effectiveness of various patient support programs• Employed Cox Proportional Hazards (CoxPH) and XGBoost as base models, achieving an R² score of 86%• Evaluated individual treatment effects at the user level, providing the flexibility to scale insights to any level Show less

• Headed a team of 24 members.