Catching Fraud Transactions in Realtime
Built Python-enabled streaming and modeling workflow support for fraud detection with low-latency response goals.
03 / Skill · Python
Python.
I build Python workflows for data preparation, pipeline automation, and analytics — code that moves from notebook to production, not just experiments.
About
I use Python as the execution layer of my analytics work: ingestion scripts, transformation logic, Dagster-orchestrated pipelines, and modeling experiments. My emphasis is reproducibility, maintainability, and measurable business outcomes — the same habits I apply to production system support at ZooTampa.
Impact Highlights
- Built a daily-refresh analytics pipeline in Dagster that joins weather API data with transaction tables for downstream BI.
- Built fraud, churn, and vision modeling workflows aligned to practical operational constraints.
- Improved automation reliability by separating experimental logic from deployable workflow steps.
Projects
Modeling Churn
Ensemble learning project that reached strong validation performance and sets up next-step generalization work.
Camera Trap Detection
Transfer-learning pipeline for wildlife imagery using pretrained models against noisy real-world data.
Tools
Python
pandas
numpy
scikit-learn
Dagster
Databricks
Jupyter
API Integrations
Automation Scripts
Process
- Anchor each build to a business problem and measurable signal.
- Establish baseline model or script behavior quickly.
- Refactor toward reusable modules and robust data contracts.
- Evaluate performance and operational tradeoffs before handoff.
- Document assumptions, known limits, and next iteration options.