📊 Data Scientist with DevOps discipline — I turn messy operational data into predictive models with measurable impact, and ship them with CI/CD, Docker, and MLOps guardrails.
carol = {
"name" : "Carollina Anderson",
"location" : "Campo Grande, MS · Brazil",
"identity" : "Data Scientist | ML + MLOps / DevOps",
"contact" : "andersoncarolinna@icloud.com",
"core" : ["Python", "SQL", "Scikit-Learn", "XGBoost", "Forecasting", "Classification"],
"mlops_devops": ["MLflow", "Databricks / Spark", "Docker", "GitHub Actions CI", "Alembic", "pytest", "AWS (S3, Athena, Glue, QuickSight, SageMaker)"],
"devops_proof": ["first GitHub repo from zero + branching/review", "internal guide trained 80+ staff", "multi-stage non-root Docker", "bandit scan + secrets + pinning"],
"domain" : ["contact-center ops", "workforce management", "retention", "routing", "customer experience"],
"proof" : ["98% -> 0.2% data quality, 20k+ docs", "42 ML projects end-to-end", "80+ staff trained via internal guide"],
"currently" : "building 2 capstones: call-volume forecasting + churn analytics",
}| Project | Business question | Result |
|---|---|---|
| call-volume-forecast (building) | How many contacts arrive next week, how many agents needed? | SARIMA/Prophet/XGB, backtest MAPE/WAPE vs naive, staffing math — repo coming |
| churn-cx-analytics (building) | Which customers will leave, and why? | LogReg vs RF vs XGBoost, ROC-AUC, SHAP, retention math — repo coming |
| law-firm-ml-case-study (sanitized, NDA-safe) | Predict delays from 20k+ processes? | RF/XGB, validation 98% -> 0.2%, no raw data — sanitized case |
| task-manager-api | How to serve models behind a tested API? | FastAPI + Postgres + Docker (multi-stage non-root) + GitHub Actions CI + bandit + Alembic + pytest — DevOps serving pattern for /predict |
NDA note: work data is private. Public repos show the same pattern on public data. I can whiteboard the private flow.
- Versioning: first repo from zero, branching/review, 80+ trained
- CI: GitHub Actions + pytest + bandit + Alembic, idempotent
--dry-runupserts - Containers: Docker multi-stage non-root, Compose for Postgres/API
- Observability: structured JSON logs, QuickSight, drift/error checks
handson-ml2·kaggle_intro_to_machine_learning·kaggle_pandas— coursework, own notebooksolist-ml-models(fork of TeoMeWhy) — studied Feature Store -> ABT -> MLflow pattern, applied privatelydata-collect— web scraping & pipelines (Python, BeautifulSoup, Requests)
⚡ Open to remote Data Scientist roles in Brazil — forecasting, classification, retention, CX analytics. 🎨 Curator / AI artist work lives on Behance.