I build intelligent systems that turn ideas into impact

A machine learning application deployed as a web app to support preliminary malaria diagnosis, built and shipped end-to-end from model to a live, usable interface.

Learned: Deploying the model taught me how much of the real work in ML applications happens after the model is trained, packaging, interface, and making it reliable for someone else to use.

AgriGuard uses fine-tuned ResNet50 models to detect diseases and pests in Cassava and Maize from leaf photos, combining confidence thresholding with energy-based out-of-distribution detection to reduce unreliable predictions. It gives treatment recommendations, preventive measures, and downloadable PDF reports, with a multilingual interface (English, Hausa, Igbo, Yoruba, Swahili, French).

Learned: Built as a team capstone for TechCrush Cohort 7 working through model validation (confidence + OOD detection) as a team sharpened how I think about making predictions trustworthy, not just accurate.

AgroPulse AI helps identify likely livestock diseases from symptoms and supports farm decision-making, built as part of an AI hackathon.

Learned: This project sharpened how I think about applying AI to agricultural problem spaces, not just conventional tech domains.

Sutton is a conversational AI bookkeeper that lets traders record business activity naturally through chat. It uses Groq-powered language models to extract debts, inventory changes, and customer information from unstructured messages, then stores those records as persistent memories using Walrus Memory. Traders can later ask questions about previous transactions and Sutton retrieves the relevant information to answer them.

Learned: Built for the Lagos Walrus Memory x AI Prompt Hackathon, this project strengthened my understanding of AI memory systems, structured information extraction, semantic retrieval, and integrating decentralized storage with conversational AI.