Dynamic Risk Assessment System
Production ML monitoring, drift detection, and retraining
Udacity MLOps system that assumes a model is already in production, then checks on a crontab for new datasets. Tests for model drift, retrains when needed, and writes performance, data-quality, and timing reports. Originally framed on the portfolio as production monitoring for all stakeholders.
The Core Challenge
Engineering Solution & Implementation
Flask-deployed pipeline using scikit-learn metrics, pandas/numpy EDA, cron for the daily job, subprocesses for CLI outputs, SciPy statistics, timeit for module timing, and pickle for model serialisation. Reports land in a local store / database.
Verified Outcomes & Deliverables
Automated drift checks and optional retraining on new data.
Persisted model-performance, data-quality, and execution-timing reports.
Shipped as part of the Udacity Machine Learning DevOps Engineer nanodegree.