- Led deployment of a production multi-agent LLM system using LangGraph and LangChain; optimized orchestration and reduced end-to-end graph latency by 50%.
- Designed tool-calling and prompt architectures and ran structured A/B experiments in LangSmith on curated evaluation datasets, resulting in higher response quality.
- Built LLM regression testing framework integrated into CI/CD, gating model updates and preventing quality degradation during rapid iteration.
- Developed internal human-in-the-loop annotation platform to capture structured feedback and accelerate model refinement cycles.
- Partnered with executive leadership to define AI roadmap and establish measurable evaluation standards for production AI systems.
Work history
- Estimated revenue and lift for new product ideas and A/B variants to guide roadmap decisions.
- Partnered with ML engineers on payment routing experiments estimated to lift yearly revenue recovery for a single customer by $750k.
- Led development of the Payment Insights Dashboard: authored PRD, delivered MVP, and demoed to clients.
- Prototyped a payment segmentation model enabling more targeted product and marketing strategies that improved campaign relevance.
- Delivered $600K+ annual cost savings via optimized billing retry logic.
- Modeled churn with gradient-boosted trees and survival analysis in Azure ML, then applied predictions to target strategic experiment cohorts, enabling data-driven product changes that reduced churn.
- Built in-house tools (sample size, estimated run time, and z/t-test calculators) to streamline and scale product A/B testing.
- Developed algorithms to detect recurring payments, flag large transactions, and track cash flow, enhancing transaction monitoring and customer insights.
- Deployed real-time drive-detection accuracy scoring and feedback pipelines on Azure, lowering detection latency and improving scoring reliability for the MileIQ product.
- Ran A/B tests for Outlook/MileIQ features and built PySpark pipelines to evaluate model performance, enabling data-driven decisions that increased feature adoption and model accuracy.
- Developed a 24-month cohort-based LTV model by extrapolating from 12-month customer purchase data, enabling finance and growth teams to generate more accurate revenue forecasts.
- Automated monthly updates with refreshed actuals to provide reliable forecasts for finance and growth teams.
- Guided revenue accounting and CAC thresholds, revealing cohort spend trends tied to pricing, fees, and regulatory shifts.
- Co-invented data security patents and developed an NLP-powered data catalog search tool, enhancing data retrieval efficiency.
- Converted credit underwriting models from SAS to R, moving from serial to parallel processing; reduced test runtime to 1/7 the original using Enterprise R and SparkR in Azure cloud.
Education
Ph.D., Cognitive Neuroscience
Northwestern University
Chicago, IL
M.S., Experimental Psychology
Northwestern University
Chicago, IL
Postdoctoral Fellowship
University of California, Davis
Patents
Search Term Extraction and Optimization from Natural Language Text Files
U.S. Patent No. 11,704,350, Jul 18, 2023
System and Method for Identifying Leaked Data and Assigning Guilt to a Suspected Leaker
U.S. Patent No. 11,350,147, May 31, 2022
Statistical Fingerprinting of Large Structured Datasets
U.S. Patent No. 11,163,745, Nov 2, 2021
Resume
Full PDF with complete details.