Looking for a job

Michael Arnold

@michaelarnold2

Senior Data Scientist at Mercury building LLM investigation tools and fraud models that reduce unnecessary reviews while maintaining fraud capture.

United States

Message

At Mercury, I prototyped an LLM-powered investigation assistant that uses semantic retrieval and RAG to surface policies and draft evidence-linked case summaries. I also built an evaluation framework from analyst-reviewed cases to assess retrieval relevance, factual consistency, citation accuracy, and appropriate abstention.

At Uber, I developed personalized restaurant-ranking models and search improvements for Uber Eats, evaluating them through A/B tests and ranking metrics. At Chewy, I owned Autoship retention experiments and developed churn-risk models to support targeted outreach.

Experience

Work history, roles, and key accomplishments

Current

Senior Data Scientist

Current

Mercury

Jan 2023 - Present (3 years 9 months)

Prototyped an LLM-powered investigation assistant using text embeddings, semantic retrieval, and RAG to surface relevant policies and draft evidence-linked case summaries. Built an LLM evaluation framework from analyst-reviewed reference cases, measuring retrieval relevance, factual consistency, citation accuracy, and appropriate abstention.

LLM RAG Text Embeddings Semantic Retrieval Python SQL XGBoost Feature Engineering Time Based Analysis Prediction Threshold Optimization Model Monitoring Drift Monitoring Feature Attribution Data Validation

Data Scientist II

Uber

Oct 2018 - Dec 2022 (4 years 2 months)

Developed personalized restaurant-ranking models with XGBoost on Michelangelo, Uber's ML platform, using order history, cuisine preferences, session context, and near-real-time restaurant features to improve discovery-to-order conversion in A/B tests. Designed and analyzed ranking experiments using CUPED variance reduction, sequential testing, and cancellation and customer-experience guardrails.

XGBoost Michelangelo A B Testing CUPED Sequential Testing Recall@K NDCG@K Spark Hive Feature Stores Embedding models Knowledge Graphs Position Bias Exposure Bias

Data Scientist

Chewy

Aug 2016 - Aug 2018 (2 years)

Owned Autoship retention A/B tests end to end, from hypothesis through measurement, reducing cancellations among treated at-risk subscribers while controlling for promotional cost, contact frequency, and cohort mix. Developed churn-risk models with Python, SQL, and scikit-learn using purchase cadence, shipment changes, and customer-service interactions to prioritize retention outreach.

Python SQL scikit learn A B Testing Churn Modeling Cohort Analysis Customer Lifetime Value Forecasting