FARUQ AFOLABI
AI EngineerNone
Faruq Afolabi is an AI Engineer at Tolaram Group Africa where he builds production AI systems across consumer brands including Indomie, Colgate, Addme, Lush Hair and Poweroil. He holds a B.Tech in Physics Electronics from the Federal University of Technology Akure .
His research spans clinical AI fairness for African populations, TinyML, multilingual African NLP and medical imaging. He has published work accepted at MICCAI, EUVIP, AIMed and ICML, with projects ranging from a 2.74MB offline crop disease detection app for smallholder farmers to a fully offline multilingual medical LLM for rural Nigerian health workers.
Faruq is the founder of AI4Africa, a practitioner community of over 800 African AI professionals, and co-founder of CongAI. He has trained over 600 students across Nigeria in AI and machine learning and runs the YouTube channel AI Research with Faruq covering applied ML and research topics for African learners.
His talk at PyCon Africa covers MobileCrop, a knowledge distillation pipeline compressing EfficientNetB0 to MobileNetV2 for deployment as a 2.74MB TFLite model on Android with no internet dependency.
Abstract
MobileCrop is a complete TinyML system for offline crop disease diagnosis targeting smallholder farmers in Sub-Saharan Africa. Using knowledge distillation from EfficientNetB0 to MobileNetV2, the student model achieves 83.81% test accuracy across 17 disease classes, marginally exceeding the teacher. Full INT8 quantization produces a 2.74 MB deployment binary for a production native Android application named Crop Doctor, built in Kotlin with TensorFlow Lite. A companion Progressive Web App delivers Float16 and Dynamic-Range TFLite variants achieving 92.67% accuracy at 2.55 MB. The poster covers the Python training pipeline, the quantization accuracy paradox where INT8 collapsed to 23.67% due to missing calibration data while Float16 gained 9 percentage points, per-class performance across three crops, and field testing results on real Nigerian farm photographs. MobileCrop is the first multi-crop TinyML diagnostic system for West African agriculture deployed as a production Android application.