Corey smiling in sunglasses and a maroon sweatshirt, standing in a field of wildflowers with a lake and mountains behind him.

Data Scientist Orlando, FL

Corey Michaud

I build models, data pipelines, and dashboards, mostly in Python and SQL.

About

I'm an analyst at Deloitte. I lead analytics for an app with over 170,000 users and build data pipelines and dashboards for healthcare clients.

Outside of work I like taking a public dataset and seeing how far a good model can go with it. Those projects end up on GitHub and Kaggle, and I'm starting a Substack to write about the models themselves.

Now
Delivery Analyst at Deloitte
Focus
Predictive modeling and analytics
Background
Statistics
Based in
Orlando, FL

Experience

Work

  1. Deloitte

    Aug 2025 – Present

    Delivery Analyst · Lake Mary, FL

    • Lead data strategy and analytics for an LLM app with 170K+ active users: SQL KPIs, Tableau dashboards, and prompt engineering.
    • Built a Python ETL pipeline that unifies 24 sources into one RAG corpus for an EHR chatbot serving 50,000+ healthcare workers.
    • Engineered a Playwright scraping pipeline that structures data on 1,000+ cars for retrieval and analytics.
    • Designed and deployed a Power BI dashboard that replaced manual surgical instrument inventory tracking.

  2. Thompson Consulting Services

    Nov 2024 – Aug 2025

    Data Administrator · Maitland, FL

    • Supported statistical analysis of FEMA disaster relief data, surfacing patterns and trends for operational reporting.
    • Automated and standardized recurring data workflows in Excel with VBA and XLOOKUP, cutting down manual processing.

Education

  1. University of Central Florida

    Aug 2020 – May 2024

    B.S. in Statistics · Orlando, FL

Projects

All repositories
Broker compensation ▲ 41%

Market analysis · Data engineering

Benefits Market Intelligence

5.1M Form 5500 records, modeled in DuckDB, showing where broker pay in U.S. employee benefits is growing and who is capturing it.

  • DuckDB
  • pandas
  • Plotly
  • Streamlit

Live dashboard Code

ROC AUCLightGBM

0.94

Classification · ML engineering

Will You Buy an EV?

Predicts EV purchases from 668K customer records. I benchmarked LightGBM, XGBoost, and CatBoost and explained the model with SHAP.

  • LightGBM
  • SHAP
  • W&B

Live demo Code

Model comparison5-fold CV F1

Classification · Imbalanced data

Predicting AIDS Deaths

Predicts patient deaths in the ACTG 175 AIDS clinical trial (2,139 patients, 3:1 class imbalance). Tuned LightGBM reached 0.93 AUC on the test set.

  • LightGBM
  • CatBoost
  • SMOTE

Code

Confusion matrixn = 53
Random forest test-set predictions. Of 25 Alzheimer's patients, 23 were predicted correctly and 2 were predicted healthy. Of 28 healthy people, 25 were predicted correctly and 3 were predicted as patients.
Pred. ADPred. healthy
AD232
Healthy325
Accuracy
0.91
Sensitivity
0.92
Specificity
0.89

Classification · Healthcare

Alzheimer's Detection

A random forest that flags Alzheimer's from handwriting, trained on 450 features from 174 people in the DARWIN dataset.

  • scikit-learn
  • pandas
  • Seaborn

Code

408 galaxy nucleilog scale

Data visualization · Weekly

TidyTuesday

My entries for TidyTuesday, a weekly community project where everyone works with the same public dataset.

  • Plotly
  • Streamlit
  • pandas

Code

Pipelineruns on a schedule

Automation · Web scraping

First Responder

Scrapes company career pages, filters new postings by keyword, and sends matches to Discord on a GitHub Actions schedule.

  • Python
  • Playwright
  • PostgreSQL

Code

Writing

Substack

  1. The Holy Trinity: XGBoost, CatBoost, LightGBM

    What are these machine learning models, how do they differ, and why are they used so much in Kaggle competitions?