Skip to content

Navigation Menu

Sign in
Sign up
@Auston-B
Auston-B
Follow

Auston Balwinski Auston-B

Applied data scientist. Deep learning on time series, causal inference, and large-scale health data. M.S. Applied Data Science, University of Michigan

Block or report Auston-B

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Auston-B /README.md

Auston Balwinski

Applied data scientist working on causal inference, machine learning at scale, model fairness, and language model evaluation. Master of Applied Data Science, University of Michigan (2026).

Selected projects

Project What it does Techniques
Sleep Health Prediction & Phenotyping Predicts sleep duration and night-to-night consistency for 45,259 Fitbit users in the NIH All of Us cohort, identifies four sleep phenotypes, and audits where model accuracy breaks down by race and age. I managed the project and authored both notebooks, owning extraction through modeling. Live dashboard · slides BigQuery at 39M-row scale, gradient boosting, KMeans phenotyping, permutation importance, subgroup fairness evaluation, Streamlit
Michigan Automatic Voter Registration Difference-in-differences study of whether Michigan's AVR law raised youth registration and turnout, using CPS microdata. Reports fragile, directionally mixed results rather than overstating them. Causal inference, two-way fixed effects, event study, clustered SEs
Teaching GPT-2 to Answer Questions Compares three ways of adapting gpt2-medium to SQuAD — prompting, QA fine-tuning, and instruction tuning — evaluated zero-shot and few-shot on token-F1. Transformers, fine-tuning, instruction tuning, LLM evaluation
Ocean Health Monitoring Predicts dissolved oxygen across 600k NOAA World Ocean Database records (R2 = 0.98) and clusters global measurements into water-mass regimes. I owned the unsupervised pipeline end to end. Clustering (KMeans, HDBSCAN, Agglomerative), PCA/t-SNE, random forests, ablation and failure analysis
Climate Change and Global Mortality Integrates WHO mortality, CRU climate, and World Bank population data into a country-year panel covering ~130 countries and a century of records, then tests how temperature relates to mortality across age groups and causes of death. Multi-source integration, entity harmonization, correlation analysis, interactive visualization

Skills

Languages — Python, SQL

ML & deep learning — PyTorch, scikit-learn, Hugging Face Transformers, gradient boosting, clustering, dimensionality reduction, fairness evaluation

Data & analysis — pandas, NumPy, statsmodels, GeoPandas, statistical modeling, causal inference, experimental design

Tools — Jupyter, Git, BigQuery, Streamlit, Matplotlib, Plotly

Contact

austonb@umich.edu · LinkedIn

Pinned Loading

  1. Michigan-AVR-Analysis Michigan-AVR-Analysis Public

    Difference-in-differences analysis of Michigan's Automatic Voter Registration law on youth registration and turnout, using CPS microdata with two-way fixed effects and an event study.

    Jupyter Notebook

  2. QA-tuning-GPT2 QA-tuning-GPT2 Public

    Three ways to turn gpt2-medium into a SQuAD question-answerer — prompting, QA fine-tuning, instruction tuning — compared zero-shot and few-shot on token-F1.

    Jupyter Notebook

  3. Ocean-Health-Monitoring Ocean-Health-Monitoring Public

    Supervised and unsupervised ML on 600k NOAA World Ocean Database records — predicting dissolved oxygen (R2=0.98) and clustering global measurements into water-mass regimes.

    Jupyter Notebook

  4. climate-change-mortality climate-change-mortality Public

    Correlation analysis of climate variables and mortality across ~130 countries and a century of records, integrating WHO mortality with World Bank/CRU climate data.

    Jupyter Notebook

  5. sleep-predict-capstone sleep-predict-capstone Public

    Predicting sleep duration and consistency from Fitbit wearables in 45,259 All of Us participants — ML models, sleep phenotypes, and subgroup fairness. SIADS 699 capstone.

    Jupyter Notebook

AltStyle によって変換されたページ (->オリジナル) /