Resume.

TechSTEPS Limited · Lagos, Nigeria

• Spearheaded the design and deployment of predictive analytics models using Python, R and AWS, proactively identifying at-risk customers and driving a 13% reduction in customer attrition.
• Supported deployment of machine learning models to Azure environments, ensuring performance monitoring and version control through MLOps practices.
• Led the development of automated reporting solutions using Power BI and SQL, reducing manual reporting effort and saving approximately 540 operational hours annually.
• Improved data accuracy and reliability across business-critical dashboards by implementing robust data quality checks, driving a 29% reduction in reporting errors and strengthening executive decision-making.

Armstrong Global Concepts LLC · Austin, TX

• Developed machine learning models with Scikit-learn and TensorFlow across 10,000,000+ customer records, uncovering insights that supported risk mitigation and customer retention initiatives.
• Analysed employee feedback using natural language processing in Python, NLTK and spaCy, identifying factors affecting workplace satisfaction and contributing to a 34% reduction in employee turnover.
• Designed and developed interactive Excel and Power BI dashboards visualising key HR and operational metrics, giving leadership the basis for data-informed retention strategies.
• Extracted, cleaned and transformed large datasets using SQL and Python, improving reporting efficiency by 41% and enhancing data reliability.
• Created reusable data pipelines with SQL and Python, improving efficiency for ongoing analysis and reporting.

• Built default prediction models over 2.2M loans issued 2007 to 2018, comparing Logistic Regression, Decision Tree, Random Forest and XGBoost.
• Focused feature selection on eliminating post-outcome leakage, excluding fields such as recovery amounts that do not exist at the lending decision and inflate accuracy while failing in production.

• Modelled employee attrition across 1,470 records and 35 features, comparing Logistic Regression, Decision Tree, Random Forest and Gradient Boosting to a best cross-validated score of 0.907.
• Addressed class imbalance with SMOTE, as roughly one in six employees left and accuracy alone would have flattered a model that predicted retention for everyone.

• Classified corporate email traffic to surface insider threat behaviour, reaching 97.6% accuracy with a voting ensemble over Logistic Regression, Decision Tree and Random Forest, benchmarked against a Keras neural network.
• Handled severe class imbalance with SMOTE, since genuine insider activity is rare by definition.

• Integrated SHAP attribution into intrusion classifications so analysts can interrogate why traffic was flagged, deployed as a Streamlit application.
• Repurposed the same attribution analysis as a leakage diagnostic, since near-perfect benchmark scores on IDS datasets typically indicate separable labels rather than model quality.