End-to-end KYC/KYB due diligence platform that generates synthetic business entity data, applies multi-dimensional risk screening, and scores entities using a Gradient Boosting ML model — visualised in a professional compliance dashboard.
This project simulates a Know Your Customer / Know Your Business (KYC/KYB) onboarding and monitoring system used by banks, fintechs, and payment platforms. It models the full due diligence lifecycle — from entity data collection and watchlist screening to ML-powered risk scoring and dashboard reporting.
Relevance: Directly mirrors KYB workflows at Tazapay (cross-border payments KYB), Deloitte USI (compliance advisory), Oracle FCCM, and BFSI onboarding teams.
kyc-kyb-risk-engine/
│
├── generate_entities.py # Synthetic entity generator (1,000 business entities)
├── train_model.py # Gradient Boosting + Random Forest ML pipeline
├── kyb_dashboard.html # Interactive KYB risk dashboard (Chart.js)
│
├── entities.csv # Raw synthetic entity dataset
├── entities_scored.csv # Enriched dataset with ML risk scores & tiers
├── model_results.json # Metrics, feature importances, confusion matrix
│
└── README.md
| Risk Category | Indicators Modelled |
|---|---|
| UBO Opacity | Undisclosed UBOs, PEP match, sanctions against beneficial owners |
| Jurisdiction Risk | Offshore incorporation (BVI, Cayman, Panama, Seychelles, etc.) |
| Sector Risk | Cryptocurrency, Arms & Defense, Gambling, Oil & Gas Trading, Mining |
| Corporate Structure | Shell company flag, complex layering, subsidiary count, layering depth |
| Watchlist Screening | World-Check hits, OFAC/UN/EU sanctions match |
| Adverse Media | Negative news flag + count of adverse media articles |
| Documentation | Incomplete KYC documents — missing UBO declaration, registration proof |
| Director Risk | Multi-directorship across 5+ entities, dormant company registration |
| Financial Opacity | Gap between declared income and annual turnover |
Primary: Gradient Boosting Classifier (300 estimators, learning rate 0.05)
Baseline: Random Forest Classifier (200 estimators)
| Feature | Description | Importance |
|---|---|---|
| UBO Opacity Score | Composite of non-disclosure + PEP + sanctions | 28.1% |
| World-Check Hit | Watchlist screening match | 19.4% |
| Alert Signal Count | Sum of all binary red flags | 15.2% |
| Structure Complexity | Shell + layering + subsidiaries composite | 12.0% |
| Jurisdiction Risk | FATF high-risk / offshore encoding | 8.9% |
| Sector Risk | High-risk industry encoding | 6.1% |
| Financial Opacity | Log of income vs. turnover gap | 3.9% |
| Adverse Media | Negative news indicator | 3.1% |
| Docs Complete | Document completeness flag | 2.0% |
| Registration Age | Company age in years | 1.1% |
| Metric | Gradient Boosting | Random Forest |
|---|---|---|
| Accuracy | 1.00 | 1.00 |
| Precision | 1.00 | 1.00 |
| Recall | 1.00 | 1.00 |
| F1 Score | 1.00 | 1.00 |
| ROC-AUC | 1.00 | 1.00 |
| 5-Fold CV AUC | 1.00 ± 0.00 | — |
⚠️ Perfect scores are expected on synthetic data. In production, KYB models typically achieve AUC ~0.78–0.88 depending on data quality and label noise.
- Live IST clock with active screening status
- 6 KPI cards — Total entities, high-risk alerts, World-Check hits, PEP flags, shell companies, model AUC
- Alerts by Sector — horizontal bar chart distinguishing high-risk vs standard sectors
- Jurisdiction Risk Distribution — doughnut chart (Offshore / Medium / Low)
- Entity Risk Tier Split — High / Medium / Low count
- KYB Red-Flag Summary — 8 flag categories with entity counts
- ML Model Performance — 6 metrics + feature importance bars
- KYB Alert Queue Table — 205 high-risk entities, filterable by jurisdiction risk, sector, flag type, and search
| Layer | Technology |
|---|---|
| Data Generation | Python · Pandas · NumPy |
| ML Model | scikit-learn (GradientBoostingClassifier, RandomForestClassifier) |
| Feature Engineering | Composite scores, log transforms, label encoding |
| Cross-Validation | StratifiedKFold (5-fold) |
| Visualization | Chart.js · HTML/CSS · Vanilla JS |
| AML Domain | KYC/KYB · Sanctions Screening · World-Check logic · UBO due diligence |
# 1. Clone the repo
git clone https://github.com/yourusername/kyc-kyb-risk-engine.git
cd kyc-kyb-risk-engine
# 2. Install dependencies
pip install pandas numpy scikit-learn
# 3. Generate synthetic entities
python generate_entities.py
# 4. Train the ML model
python train_model.py
# 5. Open the dashboard
open kyb_dashboard.htmlKishore U.
AML/KYC Compliance Analyst | Data Analytics
📱 6303308133 | Belagavi, Karnataka | Immediate Joiner
🔗 [LinkedIn] · [GitHub]
Skills demonstrated: KYC/KYB · UBO Due Diligence · Sanctions Screening · World-Check · Python · scikit-learn · Gradient Boosting · SQL · Data Visualization · Financial Crime Compliance
All data is 100% synthetic — generated programmatically with no real entity, person, or financial data. Built purely for portfolio demonstration.