At Amini, I benchmarked LightFM, GraphRAG, and Gemini LLM retrieval approaches for Barbados government policy documents, then deployed the final system in a production mobile app. I also built an automated daily pipeline for farmer survey data and designed a Grafana dashboard to track survey quality.
I contributed to training AminiFM, a time series foundation model for African environmental data, released publicly on Hugging Face. I also processed geospatial and network data for 5G RF propagation modelling using NVIDIA Aerial Omniverse Digital Twin.
For the Cocoa Disease Detection competition, I built an end-to-end computer vision pipeline that placed fourth on Zindi. I co-built CleanSight, a litter-disposal web app, and trained and deployed its real-time litter detection model; the project placed third at Deep Learning Indaba Africa.
