Machine Learning Engineer
I build end-to-end machine learning systems — from data exploration and feature engineering to model deployment and live production apps. Every project ships. No notebooks left behind.
I'm Peter Johnson, a Machine Learning engineer based in Nigeria. I specialise in building end-to-end ML systems — from raw, messy data all the way to deployed, working web applications that real users can interact with.
My approach is problem-first and analytical. Before writing a single line of model code, I study the data, understand the business context, and ask what the wrong prediction actually costs. That discipline shapes everything — from how I handle class imbalance to how I evaluate model performance beyond just accuracy.
I've shipped a semantic job recommender powered by Sentence Transformers, a solar fault detection system using XGBoost, a customer churn API built with FastAPI, and more. Every project goes to production — not just a notebook.
Currently exploring MLOps, semantic search, and applied NLP. I document my work publicly on Medium as I build, because the process matters as much as the outcome.
Real-time job recommendation system using Sentence Transformers and cosine similarity. Scrapes live listings from RemoteOK & Adzuna APIs. Semantic matching surfaces roles beyond keyword search — ML Engineer ≈ AI Researcher.
End-to-end ML system predicting customer churn, deployed as a RESTful API via FastAPI. Benchmarked Logistic Regression, Random Forest, and Gradient Boosting for real-time inference.
Hybrid ML + rule-based system for solar plant fault detection. XGBoost classifier with PR-based 3-class interpretation. Tackled 95% class imbalance. Live interactive Streamlit dashboard.
Classification model predicting asthma severity levels from patient healthcare data. Full pipeline: EDA, data cleaning, feature selection, and multi-class model evaluation.
Scraped and analysed Amazon product reviews for Sony WH-CH720N headphones using Oxylabs API. Full pipeline from raw scraping through data cleaning to complete sentiment classification.
Python application with login authentication and structured SQLite database for solar records. Clean modular architecture separating database, auth, and application layers.
Open to ML roles, research collaborations, and interesting problems. Nigeria-based and remote-friendly.