Data Scientist. Production ML, responsible AI, and LLM systems — with a growing focus on product analytics and user-facing impact.
I turn user behavior data into product decisions. From designing A/B tests that surface real signal to building models that explain churn, engagement, and retention — I work at the intersection of data and product thinking, not just data pipelines.
I've led LLM selection for a live enterprise system, owned knowledge base architecture in production, and built agentic workflows end to end. I understand the gap between a model that works in a notebook and one that holds up at scale.
I audit production ML for bias, disparate impact, and failure modes that hurt users and create risk. Using tools like SHAP and Fairlearn, I surface what models can't explain — and I've reviewed fraud and employment models and sent them back when they didn't meet the bar, before they shipped.
Designed and implemented a rigorous fairness testing suite for production ML models across fraud detection and employment use cases. Used Fairlearn and SHAP to surface disparate impact — findings resulted in models being sent back for remediation.
Built an agentic AI system to automate Walmart's internal AI governance intake process — previously handled manually by compliance SMEs. Originally developed at an internal hackathon, then extended and hardened for scale. Routes requests, assesses risk level, processes attachments, and draws from a knowledge base built by AI governance and legal experts. Under review for org-wide adoption across 1M+ associates.
Deep learning model to classify rodent cells from spaceflight vs. ground control experiments for NASA's AI4LS research program. Includes XAI layer built with Streamlit for researcher interpretability.
Automates conversion of itineraries from emails into Google Calendar events using NLP parsing. Solves a real scheduling problem end-to-end.
Streamlit app combining Hugging Face image-to-text, LangChain translation, and text-to-speech to describe images in multiple languages with audio output.
Automated financial tool using Python for forecasting and prompt engineering with Llama 2 for transaction categorization. Interactive Tableau dashboards for spending trends and forecasts.
Processed 2.5TB of behavioral data at Merkle — reduced detection lag from hours to minutes using time series forecasting.
ML pipeline comparing five algorithms across two datasets. Covers end-to-end feature engineering, model selection, and evaluation — with customer lifetime value analysis to segment high-value groups.
Built a generative AI system to predict moving vehicle paths for collision avoidance using TensorFlow, YOLOv3, and GANs.
Read Paper ↗Developed a safety device for self-defense with an AI voice assistant and automatic SOS location reporting via GPS and NLU.
Read Paper ↗Let's
Talk.
Reach out — I'd love to connect.