8+
years building analytics and ML systems
Applied AI / ML / Statistics
I build AI, ML, and statistical systems that hold up in real use.
I'm a senior data scientist with 8+ years across healthcare analytics, registry science, LLM evaluation, agentic analytics, and independent baseball modeling. My work is strongest where data quality, model behavior, and operational decisions all need to make sense at once.
1,148
urologists analyzed in a published AQUA registry study
10 / 10
published fantasy 5x5 categories led by my predictive stack
sub-4s
dashboard SLA delivered on provider-facing analytics
Selected work
Proof over posture.
LLM evaluation and agentic analytics for healthcare teams
Led model evaluation and selection for healthcare and data science workflows, balancing quality, latency, and cost instead of treating model choice like branding.
- Built a text-to-SQL assistant with RAG over schema metadata and internal documentation
- Established guardrails, logging, and evaluation loops for sensitive-data workflows
- Created internal agent tooling and trained analysts, researchers, and data scientists on effective usage
Clinical-journey modeling and fine-tuned open models
Designed a longitudinal clinical journey format and knowledge base for claims histories so downstream ML and LLM systems could reason over patient timelines with more structure and less guesswork.
- Enabled retrieval, clustering, classification, and care-gap identification from claims histories
- Fine-tuned Llama 3 8B for clinical-journey tasks
- Built task-specific datasets, evaluations, and cost estimates for larger training and inference runs
Value-based care analytics and reporting infrastructure
Led the data science workstream for a provider-facing reporting dashboard tied to advanced payment models and value-based care.
- Built Databricks ETL and measure-calculation pipelines with validation
- Delivered predictive modeling POCs with logistic regression, XGBoost, RNNs, and claim embeddings
- Built Looker models and optimized SQL to hit sub-4-second dashboard performance targets
Registry science, EHR validation, and quality measurement
Served as statistical and computational lead for the AQUA Registry and related consulting work across quality reporting, predictive modeling, and EHR extraction validation.
- Designed analyses for validating EHR extraction and NLP pipelines
- Built predictive models for recurrence and complications across urologic procedures
- Coauthored registry-focused publications on quality measurement and practice patterns
Baseball analytics lab
Where I test ideas in public.
Forecasting stack for fantasy 5x5 baseball performance
In my baseball-data project, I built a modern baseball pipeline spanning Statcast, Retrosheet, Lahman, and feature-store layers, then used it to test predictive projection systems under walk-forward evaluation.
- Promoted a leakage-safe Predictive Stack+ evaluated on 2018-2025 preseason folds
- Beat the prior Bayesian champion on overall standardized error: 0.8963 vs 0.9262
- Led all 10 published fantasy 5x5 categories on mean error across the evaluation window
Interactive bat-ball contact simulator
Built a browser-based simulator to explore how pitch shape, swing path, timing, and contact quality change batted-ball outcomes.
- Modeled pitch flight with drag, gravity, and Magnus movement
- Created a 3D bat-path and rigid-contact collision model
- Added Monte Carlo perturbations to visualize likely landing spread and contact sensitivity
How I approach the work
Applied, measurable, and domain-aware.
I care less about whether a system sounds advanced than whether it is measurable, legible, and deployable. That usually means getting the data model, evaluation frame, and workflow right before claiming a model is the interesting part.
In practice, that leads me toward strong baselines, careful error analysis, explicit tradeoffs, and human-in-the-loop designs that still make sense after the demo ends.
Publications and talks
Research, registry work, and public speaking.
Characteristics of Participants in the American Urological Association Quality (AQUA) Registry and Early Impact of Participation on Quality of Care
Coauthored publication in Urology Practice on AQUA adoption and quality-measure performance. The study analyzed more than 125 practices and 1,148 urologists.
A Snapshot of Contemporary Urologic Practice From the American Urological Association's Quality Registry (AQUA)
Coauthored Journal of Urology abstract summarizing contemporary urologic practice patterns from AQUA, spanning 200 practices, 1,731 providers, and 4.3M+ patients.
Enabling Value-Based Health Care Using Modern Analytics Tools
UMBC Data Science Meetup talk on healthcare analytics platforms, modern tooling, and the practical mechanics of value-based care data work.
Technical focus
Tools are means, not identity.
GenAI / LLM systems
Evaluation and benchmarking, RAG, tool-using agents, analytics assistants, prompt and workflow design, and fine-tuning.
Modeling and statistics
Predictive modeling, experimental design, model selection, calibration, measurement design, error analysis, and decision support.
Data platforms
Python, SQL, Spark, Snowflake, Databricks, Azure, AWS, MLflow, Tableau, Looker, and production-minded analytics workflows.
Domain experience
Healthcare claims, registry and quality analytics, real-world evidence, episode-based risk adjustment, and baseball modeling.
Contact
For serious AI / ML / analytics work.
If you need help with applied AI, healthcare analytics, statistical modeling, or decision-support systems, I'd be glad to talk.