Work
Built for real problems.
Everything here sits on one stack. The data layer: pipelines, ERP integration, and schema discovery on undocumented systems. The model layer: machine learning, analytics, and evaluation. The delivery layer: AI agents, business intelligence, and production platforms that put the numbers in front of the people deciding. Different projects enter at different layers. The goal never changes.
Data
Models
Delivery
ZimCropGuard: multi-crop disease detection under domain shift
Ongoing research: a hybrid deep-learning framework for multi-crop disease detection across real African field conditions. Best Poster, IndabaX Zimbabwe 2026.
Machine learning researchModelsTomato leaf disease detection: hybrid CNN-Transformer
A completed BSc dissertation, deployed as a live Streamlit app: a robustness-engineered hybrid CNN-Transformer that closes the lab-to-field accuracy gap on budget phones.
AI engineeringDeliveryAI agent orchestration
AI agents that query live enterprise databases in real time, wrapped in governance and safety layers, with every figure computed by deterministic code rather than the model.
Enterprise integrationDataLive ERP integration
Connecting to production ERP environments, reverse-engineering undocumented schemas, and building analytics pipelines on dirty enterprise data.
Legacy modernisationDeliveryLegacy system modernisation
Replacing fragile legacy systems with secure, modern web platforms, from undocumented databases to a single trustworthy source of record.