About Takudzwa Shumbamhini

University of Cape Town

Cape Town, South Africa

Specialisation: High Performance Computing, Embedded Systems & Computer Science

St Francis Xavier's Kutama College

Norton, Zimbabwe

Specialisation: Maths, Physics, Computer Science

Cognitive Systems

Cape Town, South Africa

• • Researched, benchmarked, and deployed multiple object-detection backbones (ConvNeXt, Faster R-CNN) for a live CCTV security platform, building a configurable model-selection layer enabling zero-downtime runtime switching.
• • Replaced a legacy HSV-histogram colour classifier with a perceptually-grounded ML pipeline (CIELAB + K-Means + CIEDE2000), improving metadata quality for downstream search and alerting features.
• • Built a behaviour analytics engine on top of object detections: implemented polygonal region-of-interest matching, contour tracking, and a loitering/directness displacement index driving a real-time rules engine for automated Telegram and cloud alert dispatch.
• • Designed tamper-detection security software for Linux-based IoT devices; migrated GPIO I/O from Arduino/Raspberry Pi to Sonoff Wi-Fi Smart Switch, improving hardware reliability by over 50%.
• • Engineered a multi-threaded event pipeline using bounded queues to decouple producers and consumers, with dedicated worker threads for persistence, rules evaluation, and hot model/rules reloading.
• • Designed and deployed a CI/CD infrastructure using Docker and Ansible, enabling reliable automated delivery of security-critical software to client environments.

Cognitive Systems

Cape Town, South Africa

• • Built a production asynchronous computer vision inference pipeline: buffered camera frames into queues, ran YOLOX/Keras models on background workers, and filtered/rescaled detections for real-time downstream alerting.
• • Designed and implemented a rule engine scorer that ingests model outputs, applies time/region/tag rules, computes per-tag confidence scores, and emits actionable alerts with scheduling and activation tracking.
• • Developed image classification and segmentation functions using MSE, SSIM, and OpenCV; implemented a correlation tracking algorithm using dlib; trained ML models using TensorFlow and Keras with YOLOv3 for object detection and annotation.
• • Established MQTT message-brokering infrastructure (Mosquitto) between IoT devices and backend systems; improved legacy codebase performance through multithreading and modularisation.
• • Built custom APIs delivering real-time AI-analysed image feeds to Telegram channels as part of a commercial security solution serving live deployments.
• • Engineered Django-based RESTful APIs enabling customer access to security event data stored in SQL databases via a web-based dashboard.
• • Mentored graduate engineers in machine learning deployment, TDD, Docker, SQL, and Python best practices.

Cognitive Systems

Cape Town, South Africa

• • Trained with senior engineers on creating project specifications and practice Agile methodologies with regards to project management and software development.
• • Improved quality image detection from video feeds Python OpenCV.
• • Built and maintained a web-based graphical user interface using the Django web framework.
• • Developed REST APIs which communicate with backend databases to serve users output from security-based software.
• • Developed, deployed and maintained Linux shell scripts to regulate software packages deployed on 100+ IoT devices.