Sep 2025 – Present · 6 mos

Sep 2021 – Oct 2025 · 4 yrs 1 mo

• Leading a team responsible for developing model governance frameworks and implementing capabilities for high-impact AI/ML algorithms across the organization.
• Key accomplishments include:
• 1) Developed standards and guidelines for hyper-parameter tuning (HPT) and model interpretability (MI),
• 2) Collaborated with Technology to implement scalable solutions, and
• 3) Conducted extensive research on: conceptualizing Model complexity and using Bayesian optimization in HPT; detecting counter-intuitive features, automating their root cause analysis, and modifying SHAP for proprietary algorithms in MI.
• Current focus is on:
• 1) Concept Drift detection for tree ensemble models,
• 2) Data sampling using modified Shapley value concepts, and
• 3) End-to-End (E2E) governance adherence capability.

Sep 2018 – Aug 2021 · 2 yrs 11 mos

• Led the Data Sciences team which created bespoke campaign audiences within People Based Marketing. PBM is a novel Amex prospecting initiative, which aims to combine disparate digital IDs into a unique People based ID to make messaging across channels more consistent and marketing more measurable.
• In parallel, led the Horizontal Data Sciences team, which used Vision & NLP techniques to create custom AI/ML solutions and enhance prospect acquisition in Paid Search & Media channels. Examples include finding next best creative for testing, evaluating quality of a mailer creative, and intelligent bidding in Paid Search.

May 2017 – Sep 2018 · 1 yr 4 mos

Mar 2016 – May 2017 · 1 yr 2 mos

• Developed AI/ML solutions and scalable products to help power Amex Advance, where we harnessed proprietary data assets and capabilities to create predictive audiences & exclusive insights for brands, advertising networks & publishers.
• This required us to create custom audience models for partners, where we leveraged NLP for some use-cases to learn user intent from browsing data (topic modelling) and for merchant aggregation into brands (entity matching).
• Eventually, we also automated our solutions into products to help partners distill browsing data into relevant audience characteristics, and an E2E modelling pipeline for high volume audience users.