I Transform Data into Strategic Actionable Intelligence

In large-scale agent banking and fintech networks, compliance teams manually tracking fraud and operational irregularities face massive scalability bottlenecks. This project simulates an expanding financial ecosystem processing over 500,000 transactions across 5,000 active field agents. Without automation, identifying regional risk hotspots, tracking volatile service provider failure rates, and catching bad actors happens reactively—often after significant capital or reputational damage has occurred.

My objective was to design, develop, and deploy an end-to-end data intelligence and risk-scoring framework. The system needed to automatically ingest raw operational data, enforce strict relational data integrity rules, run predictive compliance algorithms, and surface actionable, real-time priorities directly to field auditors and C-suite risk executives.

I engineered a modular data pipeline split into three distinct core architectural layers:

• Data Engineering (Python): Architected a 7-stage data processing pipeline handling generation, cleaning, automated missing-value logic, and error handling.
• Database Optimization (SQL): Built a production-grade relational schema (schema.sql) enforcing primary/foreign keys and specialized analytical queries.
• Predictive Risk Scoring (ML): Deployed predictive algorithms evaluating documentation quality, historic fraud flags, and incident volumes to calculate active risk scores (≥ 70 probability).

Delivered a functional, terminal-driven data pipeline paired with a 4-page interactive Power BI workspace. The system eliminates manual auditing overhead by instantly transforming backend transaction logs into evidence-based field intervention plans, mapping macro performance vectors and geopolitical threat shifts in real time.

In public finance and macroeconomic auditing, manually tracking multi-year compliance shortfalls and structural revenue losses across shifting fiscal calendar boundaries introduces severe calculation bottlenecks. Analyzing multi-billion tax gap figures across a matrix of publication cycles (2014–2027) without automation causes reactive identification of drivers like VAT or Corporation Tax, obscuring long-term revenue loss trends.

My objective was to design, optimize, and deploy a robust relational reporting model and executive BI application. The project required ingesting multi-stream transactional structures, resolving evaluation context mismatches between dimensional calendars and fact metrics, and surfacing evidence-based revenue insights via dynamic KPI formatting indicators.

Engineered a unified corporate analytics workspace across three primary layers:

• Data Integration (Power Query/M): Applied column-level filtering and type-safe casting wrappers to reconcile fiscal year cycles.
• Context-Aware Modeling (DAX): Neutralized visual context errors using REMOVEFILTERS and COALESCE fallback logic.
• Dynamic SVG UI Engineering: Programmed inline XML vector calculations (data:image/svg+xml) that inject color-coded trend indicators dynamically.

Stabilized metric engine guaranteeing 100% calculation accuracy whether data is filtered or viewed unfiltered. Deployed an executive intelligence suite with macro variance matrices and synchronized KPI cards, instantly surfacing compliance drivers (a +27.6% / £4.22B total surge driven by VAT Retail & Wholesale and Large Business CT) to inform legislative interventions.

NHS hospital pharmacy networks faced persistent supply chain vulnerabilities, including unpredictable demand spikes for critical medicines (e.g., Antibiotics, Analgesics), stockout risks, overstocking costs, and raw data anomalies like negative stock records and missing values.

Develop a robust, automated end-to-end Python engine to preprocess hospital clinical activity data, evaluate predictive models (SARIMAX vs XGBoost), set up threshold alerts based on Business Rules (BR-002, BR-003), and deliver an executive dashboard in Plotly/Dash.

Executed technical architecture across feature engineering, predictive modeling, and dashboard development:

• ETL & Preprocessing: Handled stock anomalies, forward-filled missing values, merged clinical & activity records, and extracted 1/3/6/12 lag features plus 3-month rolling average.
• Dual Model Engine: Trained SARIMAX using hospital bed occupancy rates as exogenous factors and XGBoost Regressors with hyperparameter tuning.
• Alert Logic: Implemented rules flagging critical stockout risks (<30 days coverage) and overstocking (>120% demand ceiling).

Reduced demand forecast error (MAPE) from 6.4% (SARIMAX) down to 3.1% (XGBoost). The system successfully monitored 45 SKUs across 8 NHS hospital facilities, highlighting 199 urgent reorder requirements and providing real-time inventory oversight.