Dear Hiring Manager,
Please find below a summary of the Prime Capital Bank Data Intelligence Platform — a portfolio project
I have designed and built end-to-end to demonstrate production-grade data engineering and machine-learning
skills in an enterprise banking context. Full source code, notebooks, and documentation are available on
GitHub (link below) for your review.
Platform Overview
Prime Capital Bank is a fictional Fortune 500 South African retail and corporate bank with
500,000 active customers, a R2.3 trillion asset base, and a 200-branch network.
The platform consolidates twelve previously siloed source systems — Temenos T24 core banking,
Salesforce CRM, Kondor+ treasury, Oracle Financials general ledger, and third-party feeds from SARB,
TransUnion, and Reuters — into a single governed, cloud-native data intelligence platform on
Azure Databricks with a Delta Lake medallion architecture.
Azure Databricks
Delta Lake
MLflow
Azure Data Factory
Power BI
hive_metastore
LightGBM / XGBoost
PySpark
dbt Core
SARB / IFRS 9 / Basel III
Deliverables Included in This Submission
📖 Technical Reference Ebook
12-chapter HTML ebook covering architecture, all medallion layers, 12 ML models, SA fintech ecosystem (Yoco, SnapScan, PayFast, Peach Payments, PayGate), regulatory compliance, and data governance. Includes Chart.js visualisations.
📊 Excel Analytics Workbook
Dynamic Excel workbook with live formulas, conditional formatting, and KPI dashboards covering credit risk, fraud metrics, AML exposure, regulatory capital (BA700), fintech fee revenue, and pipeline SLA performance.
🗄️ Interactive ERD + Data Dictionary
6-tab Mermaid.js ERD (zoom/pan) across all Gold schema domains including Cards & Fintech. Companion searchable data dictionary covering all 15 Gold tables with PK/FK badges, data types, and business descriptions.
🗺️ Merchant Intelligence Map
Interactive Leaflet.js map of South Africa showing merchant transaction volumes (circle size = sales) and fraud/AML hotspots across 9 provinces. Toggle between Sales Volume, Fraud Heatmap, and AML Case views.
⚙️ Full Source Code (GitHub)
9 Databricks notebooks (Setup → Bronze → Silver → Gold → 12 ML models → Regulatory → Orchestrator), dbt transformations (staging + intermediate + marts), 4 Gold analytical views, SQL DDL/DML/seed scripts, Python data generator (12M+ rows, 9 CSVs), Terraform IaC, and Azure provisioning scripts.
Technical Ebook — Screenshots
PRIME CAPITAL BANK
Data Intelligence Platform
Azure Databricks · Medallion Architecture · ML & Analytics
Ebook Cover — Prime Capital Bank Data Intelligence Platform (12-chapter Technical Reference)
CHAPTER 7
ML Models
PD Model (Basel IRB)
0.94
Gini Coefficient
LGD Model (IFRS 9)
0.89
R² Score
Fraud Detection
0.97
AUC-ROC
Ch 7: ML Models — PD, LGD, and Fraud model performance metrics
CHAPTER 9
Pipeline Orchestration
Databricks Workflows · 9-task DAG · Medallion Architecture
| Pipeline |
Schedule |
SLA |
| Fraud Streaming |
Continuous |
<50ms |
| Bronze — Cards |
Every 5 min |
T+10min |
| Bronze — Core Banking |
Hourly |
T+45min |
| Gold Star Schema |
Daily 02:00 |
T+4h |
Ch 9: Pipeline Orchestration — Databricks Workflows scheduling hierarchy
Enterprise Data Model ERD — Screenshots
PRIME CAPITAL BANK
Complete Enterprise Data Model
Production-Ready ERD — Azure Databricks Unity Catalog | Delta Lake | Medallion Architecture
22 Dimensions · 16 Fact Tables · 60+ Products · IFRS 9 / Basel III / SARB / POPIA
ERD Header — 38 entities across 6 domain tabs with Medallion data flow diagram
Entity Relationship Diagrams — by Domain
① Customer & Accounts
② Lending & Risk
③ Products & Channels
④ Cards & Fintech
⑤ Finance & Compliance
⑥ Insurance & Investments
Core customer identity, account ownership, and daily transaction events. DIM_CUSTOMER is SCD Type 2 tracking historical segments and risk scores.
ERD Domain Tabs — 6 light-theme diagrams, 7–9 entities each
Medallion Architecture — Data Flow
SOURCE
Core Banking
CRM · Cards
Treasury · GL
→
BRONZE
Raw Ingest
Auto Loader
14 Tables
→
SILVER
Cleansed
PII Masked
11 Tables
→
GOLD
Star Schema
22 Dims
16 Facts
→
ML LAYER
MLflow
PD·LGD·Fraud
AML
Medallion Architecture Data Flow — Bronze → Silver → Gold → ML → Consumption
ML Model Performance Metrics
0.94
PD Model — Gini
XGBoost · Basel IRB · 5-fold CV · SMOTE · SHAP
0.89
LGD Model — R²
Gradient Boosting · IFRS 9 ECL dependency · Collateral-aware
0.97
Fraud Detection — AUC-ROC
GBM + Isolation Forest ensemble · Real-time <50ms scoring
PSI <0.1
Model Drift — Stable
Population Stability Index · MLflow drift monitoring · Quarterly reviews
Azure Databricks — Pipeline Architecture
🔷 Live Azure Databricks Workspace:
The pipeline is deployed and has been executed on Azure Databricks at:
https://adb-7405610635095106.6.azuredatabricks.net
The 9-task Databricks Workflow DAG (Job ID: 303640560596013) runs the full medallion pipeline:
Setup → Bronze Ingestion → Silver Cleanse → Gold Star Schema → Credit Scoring (ML) →
Fraud Detection → AML Risk Scoring → Regulatory Reports → Orchestrator Summary.
All notebooks are version-controlled in GitHub. Log in with Microsoft Entra ID credentials
to explore the live workspace, Delta tables, MLflow experiments, and job run history.
| # |
Task |
Notebook |
Layer |
Duration |
| 1 | workspace_setup | 00_workspace_setup.py | – | ~2 min |
| 2 | bronze_ingestion | 01_bronze_ingestion.py | Bronze | ~8 min |
| 3 | silver_cleanse | 02_silver_cleanse.py | Silver | ~12 min |
| 4 | gold_build | 03_gold_star_schema.py | Gold | ~25 min |
| 5 | credit_scoring | 04_ml_credit_scoring.py | ML | ~18 min |
| 6 | fraud_detection | 05_fraud_detection.py | ML | ~22 min |
| 7 | aml_risk_scoring | 06_aml_risk_scoring.py | ML | ~30 min |
| 8 | regulatory_reports | 07_regulatory_reporting.py | Regulatory | ~10 min |
| 9 | orchestrator_summary | 08_orchestrator.py | – | ~3 min |
Source Code — GitHub Repository
⬡
Full Source Code — Review on GitHub
https://github.com/anthonyapollis/prime-capital-bank
9 Databricks notebooks · ERD · Ebook · Excel workbook · dbt transformations · MLflow configs
The repository contains all source notebooks in /notebooks/, the technical ebook and ERD
in /docs/, the Excel workbook in /reports/, and a complete
README.md with architecture diagrams and setup instructions. All code follows
production standards: parameterised configurations, Delta Lake ACID transactions, MLflow experiment
tracking, and zero hardcoded credentials (Databricks Secret Scope throughout).
Why This Platform Demonstrates Production Readiness
This project deliberately targets the full stack that enterprise data teams operate in South African
banking: SARB POPIA compliance (PII masking in the Silver layer),
IFRS 9 ECL (three-stage loan provisioning in the Gold layer),
Basel III BA700/DI200 regulatory returns (automated in the Regulatory notebook),
and real-time fraud scoring via Structured Streaming with Kafka/Event Hub integration.
The 12 ML models — including PD, LGD, Fraud Ensemble, and AML GraphX — are each registered in the
MLflow Model Registry with drift monitoring (PSI), backtesting, and champion-challenger
evaluation built in from day one.
I would welcome the opportunity to walk through any section of the code, discuss architectural
trade-offs, or demonstrate a live notebook run. Please do not hesitate to reach out at the
contact details below.
Thank you for your time and consideration. I look forward to hearing from you.
Anthony Apollis
Data Engineer | Analytics Engineer | ML Platform