Hi, I'm Shedrack Eze
Machine Learning Engineer and MS EAI candidate at Carnegie Mellon University Africa. I build ML surrogates for solar PV systems, agentic AI tools, and open-source projects.
About Me
ML Engineer pursuing an M.S. in Engineering AI at CMU-Africa, researching ML surrogates for solar PV system design in West Africa.
I'm a Machine Learning Engineer with a background in Agricultural & Environmental Engineering (First Class Honours, FUTA). I'm currently a Research Assistant at the DEAL Lab, where I built a PV performance modeling pipeline that validated NASA POWER, ERA5, and MERRA-2 irradiance against 23-station WAPP ground data (R² = 0.974) and decomposed a 59% PV performance gap into soiling and storage effects.
Before grad school, I was Senior Data Analyst at Fluna, where I drove 60% business growth in 18 months through data mining pipelines that expanded customer acquisition by 300+ companies. I care about building things that work in the real world, especially for Africa's energy and agriculture sectors.
Skills
Projects
Selected work: research, engineering, and open-source. More on my GitHub.
ML Surrogates for PV System Design
DEAL Lab research: physics-based PV-plus-storage model under dust for West Africa. Validated irradiance products against 23-station WAPP ground data (R² = 0.974); decomposed a 59% performance gap into soiling and storage effects.
Fluna: Agri-Export Data Platform
Built customer databases and data mining pipelines that expanded customer acquisition by 300+ companies and drove 60% business growth in 18 months. A/B testing campaigns, CRM integration, revenue up 30% YoY.
MetalForge: Local Video Generation
Native Metal inference engine for AI video models on Apple Silicon (H3, Wan 2.2, LTX-2.5, Wan-Animate-2), built from scratch in C + Metal, no PyTorch/MPS dependency, with a local web UI.
Zindi Competitions
Competitive ML on African datasets: fine-tuned BERT to 90.7% accuracy, VLM fine-tuning for ASR, ensemble methods. Top-tier leaderboard results across multiple challenges.
Coffee Pricing Model
Automated daily coffee futures pricing signal with dual-basis analysis (stale vs current basis), live data pipeline, and a 2-page decision brief. Built for Fluna's export operations.
PV Soiling & Storage Analysis
Soiling loss quantification across West African climate zones using MERRA-2 aerosol data and Koppen-Geiger classification; battery state-of-charge validation with three independent estimators.
Movie Sentiment Analysis
Fine-tuned BERT on 50,000 IMDB reviews to 90.71% accuracy, overcoming GPU memory limits with gradient accumulation and mixed-precision training. Deployed a live app on Hugging Face Spaces. Live demo
Dog Breed Image Classifier
Compared AlexNet, ResNet, and VGG for 133-class identification with transfer learning; achieved optimal accuracy with VGG and reduced training time by 70% via ImageNet pretraining.
Certifications
Professional credentials in cloud, ML, and data analytics.
Get In Touch
I'm open to research collaborations, ML engineering roles, and interesting problems, especially in energy, agriculture, and AI for Africa.