How priyanka-senthil/mlops_project_financialcrises works
The Financial Stress Test Platform turns economic and company data into scenario-based forecasts and risk assessments, served through an authenticated API. Its principal user flow is company and scenario selection, feature mapping, three-model inference, and explained risk results. Separate training and data-preparation workflows support the serving path; monitoring, drift detection, alerts, and retraining are additional capabilities. The README documents the broad architecture; source-index references and excerpts support only some implementation-level links, so unsampled connections are not asserted.
Main entry points: stress_test_pipeline.py Stress-test pipeline · prediction_model_training_pipeline.py Training orchestration · step2_feature_engineering.py Feature engineering · main.py API entry point
Components 25
Data preparation
- Data collection
src/data/step0_data_collection.py - Data cleaning
src/data/step1_data_cleaning_and_merging.py - Feature engineering
src/data/step2_feature_engineering.py - Bias checks
src/data/step3_bias_detection_with_explicit_slicing.py - Anomaly checks
src/data/step4_anomaly_detection.py - Drift checks
src/data/step5_drift_detection.py - Data validation
src/validation/robust_validator.py
Model training
- Training orchestration
dags/prediction_model_training_pipeline.py - Scenario generator
src/models/Dense_VAE_optimized_mlflow_updated.py - Financial predictors
src/models/predictor/predictor_model.py - Anomaly risk model
src/models/train_anomaly_detection.py - Experiment tracking
src/utils/mlflow_config.py
Prediction service
- API entry point
deployment/api/main.py - Authentication
deployment/api/auth_system.py - Company data fetcher
deployment/api/gcs_data_fetcher.py - Model loader
deployment/api/model_loader.py - Feature mapper
deployment/api/feature_mapper.py - Stress-test pipeline
deployment/api/stress_test_pipeline.py - Risk explanations
deployment/api/stress_test_pipeline.py - Risk results
deployment/api/production_features.py
Monitoring and operations
- Monitoring dashboard
model_monitoring_dashboard.py - Alerts
src/monitoring/alerting.py - Data pipeline orchestration
dags/financial_crisis_pipeline.py
Other components
- Risk analyst
- Google Cloud Storage
Connections 27
- Risk analyst → API entry point: requests analysis
- API entry point → Authentication: authenticates
deployment/api/main.py - API entry point → Company data fetcher: fetches company data
- API entry point → Model loader: loads models
- API entry point → Stress-test pipeline: runs stress test
- Stress-test pipeline → Feature mapper: maps features
deployment/api/stress_test_pipeline.py - Stress-test pipeline → Scenario generator: generates scenarios
- Stress-test pipeline → Financial predictors: forecasts finances
- Stress-test pipeline → Anomaly risk model: assesses risk
- Stress-test pipeline → Risk explanations: explains risk
deployment/api/stress_test_pipeline.py - Stress-test pipeline → Risk results: returns assessment
deployment/api/production_features.py - Data collection → Data cleaning: precedes cleaning
- Data cleaning → Feature engineering: precedes engineering
- Feature engineering → Data validation: validates data
- Feature engineering → Bias checks: checks bias
- Feature engineering → Anomaly checks: checks anomalies
- Feature engineering → Drift checks: checks drift
- Data pipeline orchestration → Data collection: orchestrates data pipeline
- Data pipeline orchestration → Google Cloud Storage: stores processed data
- Training orchestration → Financial predictors: trains predictors
- Training orchestration → Scenario generator: trains scenario model
- Training orchestration → Anomaly risk model: trains risk model
- Training orchestration → Experiment tracking: tracks experiments
- Drift checks → Alerts: notifies drift
- Monitoring dashboard → Drift checks: visualizes monitoring
- Google Cloud Storage → Model loader: provides model artifacts
- Training orchestration → Google Cloud Storage: stores trained models