Double-level solution for stroke detection on CT scan images. Combined a DL U-Net with pretrained classification layers for domain adaptation to segment the four vascular brain areas. Applied Chan-Vese algorithm to detect hypodensity zones and automate diagnosis. Proved efficiency under low-resource dataset constraints.

Series of projects using librosa and soundfile for voice feature extraction targeting improvements in ASR and voice synthesis. Explored encoder-based architectures, CTC, and attention mechanisms. Optimized for low-resource GPU environments.

RAG-based system answering citizen questions about public services in Benin. Built for real administrative needs with accessible, conversational AI serving the general public.

Neural machine translation model built with fairseq for French → Fongbe translation. Contributing to NLP accessibility for a low-resource African language spoken in Benin.

Flask + TensorFlow platform for cancer-stage classification and early-stage prediction, designed for clinical decision support.

Fine-grained classification across 37 species of dogs and cats. Achieved over 85% accuracy with loss below 0.5 through careful model tuning and data augmentation.

Computer vision system to translate African sign language into text and audio output, bridging communication accessibility for sign language users.

Agentic RAG combining medical insurance premium regression with intelligent advisory dialogue. Users can predict their insurance costs and receive contextual advice.

RAG-based sales chatbot built with teammates at the Deep Learning Indaba Ghana hackathon. Answers product and sales questions in a conversational interface.

Object detection model to identify trash bins by color type, enabling a robot to sort waste correctly. Team project combining CV and robotics integration.