MLOps · AI Engineering
Model Release Observatory
Model Release Observatory provides experiment tracking, release evidence, and drift signals. A trained model is not yet a release: teams need to know which data produced it, whether it passed policy thresholds, and how it behaves once served.
Features
- Training runs tracked in MLflow with parameters, metrics, dataset fingerprint and model artifact
- Run comparison and a threshold-based release report (F1 and recall of at least 0.95)
- Statistical drift check using the Kolmogorov-Smirnov test
- FastAPI inference endpoint and a latency benchmark reporting P50, P90 and P99
- Reports stored as inspectable JSON files
Technologies
Pythonscikit-learnMLflowSciPypandasFastAPIpytest