Selected systems / 2023—present

The
work.

Not a gallery of mockups. This archive tracks the problem, system architecture, deployment environment and the trade-offs that mattered.

PROJECT / 01Flagship · deployed

J7 Vireo

An offline wearable assistive-vision platform that detects everyday objects, estimates distance and communicates through speech and haptic feedback.

YOLOv8mHailo-8LRaspberry Pi 5ESP32-C6
PROJECT / 02On-device · deployed

RoverCam QNN

An eight-class YOLO26m detector trained locally, converted through Qualcomm QNN and deployed to the Snapdragon Hexagon HTP in a Galaxy S24 Ultra.

YOLO26mQualcomm QNNHexagon HTPAndroid
PROJECT / 03Active · competition

Cars4Mars Rover

Six-wheel rover with four-wheel steering, Webots simulation, multi-camera perception and ordered target approach while avoiding obstacles.

WebotsYOLOAckermannControl systems
PROJECT / 04Research prototype

Road Risk Atlas

A South African crash-risk pipeline that combines location, hour, weather, calendar effects and road context to identify dangerous segments.

PythonXGBoostTime seriesGeospatial
PROJECT / 05Product concept

WhatsApp Ops

A lightweight order and invoice manager for small South African businesses, designed around the messaging workflow they already use.

Next.jsFastAPISupabaseWhatsApp API
PROJECT / 06Local-first

Local Agent

A private local assistant built around Ollama with controlled file operations, document reading and structured terminal workflows.

PythonOllamaQwenTool use
PROJECT / 07ML pipeline

Three-Class Detector

Dataset cleaning, annotation, training and deployment for hammer, traffic-cone and tennis-ball detection in the rover environment.

YOLO26mAnnotationEvaluationEdge deployment

See code, not claims.

Open GitHub