Machine Learning Engineer | NLP & Financial Services 📍 Lagos, Nigeria
I build end-to-end ML pipelines across NLP and financial services — from feature engineering and model development to deployment. My focus is production-ready work, not just notebooks.
Four-language NLP system benchmarking classical and deep learning models on low-resource African languages
- Compared Logistic Regression, SimpleRNN, LSTM, and AfroXLMR across Hausa, Igbo, Yorùbá, and Nigerian Pidgin — languages with scarce labelled data and no dominant pretrained baseline
- Champion model: AfroXLMR — F1: 0.74
- Diagnosed and resolved train/test leakage and Pidgin neutral-class collapse caused by data scarcity
- Deployed live demo to Hugging Face Spaces
- Stack: Python • PyTorch • HuggingFace Transformers • LSTM • Scikit-learn
👉 View Repository | 🤗 Live Demo
Production ML pipeline predicting loan default probability for digital lenders
- Engineered 22 credit risk features in PostgreSQL simulating a Nigerian fintech feature store (mapped to Mono/Okra, CRC bureau, CBN DTI guidelines)
- Trained and tuned Logistic Regression, Random Forest, and XGBoost using scikit-learn pipelines
- Champion model: XGBoost — AUC 0.7223
- Stack: Python • PostgreSQL • scikit-learn • XGBoost
Real-time transaction fraud detection pipeline on the IEEE-CIS dataset
- MLP classifier built in PyTorch, served via FastAPI, deployed to AWS Lambda
- Experiment tracking with MLflow
- Stack: Python • PyTorch • FastAPI • AWS Lambda • MLflow
End-to-end regression project on Nigerian real estate data. Focused on multicollinearity, OLS and regularization techniques.
ML & Modeling Python • PyTorch • scikit-learn • XGBoost • HuggingFace Transformers • Pandas • NumPy
Deployment & MLOps FastAPI • Docker • AWS (Lambda • Elastic Beanstalk • S3) • MLflow
NLP Transformers • AfroXLMR • LSTM • Multilingual NLP
Data Engineering PostgreSQL • SQL • Feature Engineering
Visualization Matplotlib • Seaborn • Tableau