Digital Payments Risk & Compliance Engine

UPI / NEFT / RTGS Flow Simulation • STR Rules Engine • ML Risk Scoring • SHAP Explainability

A Supervisory-Grade Fintech Compliance Stack Inspired by RBI DPSS Standards

---

⚡ Executive Overview

This system is a full-stack digital payments compliance engine designed to mirror the internal tooling used by:

• RBI – Department of Payment & Settlement Systems (DPSS)
• RBI – Financial Intelligence & AML Supervision Units
• Leading UPI / RTGS fintech risk divisions
• Bank AML Monitoring Cells

It simulates national-scale UPI/NEFT/RTGS payment flows, performs deterministic STR rule screening, applies machine-learning risk scoring using XGBoost, and generates SHAP-based explainability outputs required for regulatory-grade auditability.

This repository is engineered to resemble a production-aligned compliance prototype — not a classroom exercise.

---

🏗 System Architecture

Digital-Payments-Compliance-Engine/
│── data/
│   ├── audit_summary.csv
│   ├── str_audit.jsonl
│   ├── ml_upgrade_preds.csv
│   ├── ensemble_preds.csv
│   ├── shap_explanations.csv
│── src/
│   ├── simulation.py
│   ├── simple_transaction.py
│   ├── str_engine.py
│   ├── ml_upgrade.py               ← ML pipeline (XGBoost + SMOTEENN + temporal split)
│   ├── shap_explain.py             ← SHAP explainability engine
│   ├── app.py                      ← Streamlit Compliance Dashboard
│── reports/
│── plots/
│── models/
│── README.md
│── requirements.txt

---

🔁 End-to-End Flow Overview

1) Payment Simulation Layer

Generates realistic synthetic UPI / NEFT / RTGS transactions with:

• device fingerprints and IP metadata
• KYC verification attributes
• behavioural anomalies
• dormant account patterns
• reversal loops
• high-value RTGS corridors
• time-of-day patterns

Used to create an AML-scale transaction monitoring dataset.

---

2) Deterministic STR Rules Engine (RBI-Inspired)

Each transaction is evaluated through RBI-aligned AML rules:

Rule	Description
High Value	> ₹2,00,000 or abnormal UPI volume spikes
Structuring	Split transactions to avoid reporting thresholds
Velocity	Rapid transfers in very short time windows
Dormant Spike	Sudden activity after long dormancy
Round Numbers	10k / 50k / 1L pattern detection
Reversal Loop	Pay → reverse cycles
KYC Mismatch	Risk tier mismatch vs transaction amount
Time-of-Day	Night-time / unusual behavioural patterns

All rules produce risk scores, flags, and are logged to audit JSONL.

---

3) Machine Learning Upgrade (XGBoost + SMOTEENN)

A production-style ML pipeline:

• Temporal Train/Test Split (prevents leakage)
• SMOTEENN for class balancing
• RandomizedSearchCV hyperparameter tuning
• XGBoost (multi:softprob) for multi-class risk scoring
• Threshold Optimization to maximize STR recall
• Ensemble Risk Score blending deterministic + ML outputs

Produces artifacts:

• ml_upgrade_model.joblib
• ml_upgrade_preds.csv
• ensemble_preds.csv

---

4) SHAP Explainability Layer

Regulatory-grade explainability delivering:

• shap_explanations.csv (per-transaction feature contributions)
• shap_summary.png (global importance visualization)

Supports model governance and audit-driven transparency.

---

5) Streamlit Compliance Dashboard

Analyst-friendly UI for:

• filtering by flags & risk tier
• viewing STR candidates
• inspecting ML + deterministic signals
• checking SHAP explanations
• reviewing audit logs
• transaction-level drill-down

Run with:

streamlit run src/app.py

---

📊 Outputs Generated

• audit_summary.csv → All transactions + rule hits
• str_audit.jsonl → All STR flags emitted
• ml_upgrade_preds.csv → ML predictions
• ensemble_preds.csv → Deterministic + ML blended scores
• shap_explanations.csv → Per-transaction SHAP contributions
• plots/ → SHAP summary, ROC/PR curves
• reports/ → Compliance PDF-ready analysis

---

🔧 How to Run the Engine

1) Install Dependencies

pip install -r requirements.txt

2) Run Simulation

python src/simulation.py

3) Run Deterministic STR Engine

python src/run_str_engine.py

4) Train the ML Model

python src/ml_upgrade.py

5) Generate Explainability (SHAP)

python -m src.shap_explain

6) Launch Dashboard

streamlit run src/app.py

---

🎯 Policy & Regulatory Relevance

This prototype aligns with principles used by RBI:

• AML/CFT suspicious pattern identification
• Supervised ML model governance expectations
• Payment system oversight (DPSS)
• Risk-based reporting thresholds
• Operational/behavioural anomaly detection
• Data lineage & auditability (JSONL + SHAP)

It demonstrates how transaction monitoring systems integrate deterministic rule-based methods with machine learning to produce reliable STR escalations.

---

📌 Future Enhancements

• Graph-based entity linkage & network risk mapping
• LSTM sequence modelling for behavioural drift
• Real-time Kafka event ingestion simulation
• Automated SAR/STR drafting module

---

📣 Author

Designed and engineered by Naman Narendra Choudhary — fintech, ML, compliance, and engineering hybrid.

For collaborations or inquiries: connect via GitHub or LinkedIn.