Resume

• Sole engineer for a predictive/AI model discrimination testing platform. Reviewed legislation in 8 jurisdictions, iterated on business requirements, and deployed to production on Databricks in 10 weeks, owning every phase of the software development life cycle and saving a leadership-estimated ~2,500 analyst hours annually.
• Built the platform in Python (Litestar, Pydantic, ~20 REST endpoints) with a React frontend, applying software development standards (CI/CD, typing/linting, unit tests) to deliver maintainable software (~10k LOC).
• Implemented 12 statistical algorithms to test predictive/AI models for discrimination, leveraging PySpark to run large-scale analytics over 100M+ row datasets in minutes.
• Advised non-technical leadership and consultants from KPMG on the firm's global bias testing guidelines, identifying a key gap for continuous model outputs and suggesting the replacement methodology adopted into corporate policy.
• Drove adoption across a 70-person global data science organization for compliance: ran 20+ discovery interviews across seven business units, engineered authorization scheme to comply with data privacy requirements which unblocked EU adoption, and delivered an hour-long training to model owners, actuaries, and three regional VPs of AI.

Read the case study

• Developed and validated a multiclass XGBoost classifier in Python on 3M+ oil samples using stratified cross-validation, improving absolute F1 score by 15 points over the existing rules-based system and increasing diagnostic efficiency at 70+ fluid analysis labs worldwide.
• Prototyped a generative AI application to summarize vehicle inspections using open-source LLMs (LLaMA) on AWS EC2 and presented the solution to Caterpillar's Analytics and AI Center of Excellence.

• Built user interface components for Caterpillar's flagship fleet management application and optimized PostgreSQL queries for an enterprise database restructuring project.

• Designed a novel explainable AI (XAI) algorithm to interpret black-box machine learning predictions, reducing variability in local explanations by up to 5x.
• Engineered high-performance computing GPU simulations of silicon grain boundaries to guide the manufacturing of next-generation thermoelectric materials for space power generation.
• Analyzed e-commerce algorithms using statistical machine learning to optimize free shipping thresholds. Studied the ability of LLMs to generate ordering policies for inventory management problems.
• Created educational resources to teach queueing theory and inventory management to Kellogg MBA students.

• Designed and taught 8 lab sessions for the MS in Machine Learning and Data Science program covering regression, neural networks, tree ensembles, and model interpretation.
• Designed and taught a three-session, six-hour statistics bootcamp for the incoming IEMS PhD cohort.