Mosul Solutions — Banking-scale Cybersecurity Anomaly Detection (Enhanced)

This enhanced prototype includes:

• Training pipeline (PyTorch) that logs to MLflow (local mlruns/).
• Model service (FastAPI) exposing Prometheus metrics and model inference API.
• Stream processor and data generator (demo).
• Streamlit SOC dashboard that reads MLflow run metrics and Redis alerts (demo).

Quickstart:

1. pip install -r requirements.txt
2. python services/training/train.py
3. python services/model_service/app.py
4. streamlit run services/dashboard/app.py

Explainability

Included a SHAP/permutation-based explainability helper and Streamlit integration at services/explainability/shap_helper.py. The dashboard will attempt to compute SHAP via KernelExplainer if shap is installed; otherwise it shows permutation importances.