Hire this AI Trainer
Sign in or create an account to invite AI Trainers to your job.
Roboflow
Document
Image
Audio
Data Collection
Segmentation
Classification
Bounding Box
Fine-tuningI have undergone trained courses in machine learning and deep learning, during which I gained exposure to state-of-the-art AI-driven technologies such as DNNs, CNNs and SNNs. I undertook a design project which involved the development of an audio classifier using the Colab online platform. The development process required the definition and design of an appropriate model architecture, followed by annotation and training using a provided dataset of audio samples. A CNN architecture was adopted, and the model was capable of classifying unseen samples with an accuracy of 75%. The dataset was taken from a medical domain, with samples carrying indications of particular medical conditions. My master's level project involved the development of a real-time object detection pipeline. The work package included manual data collection, annotation, training, and real-time implementation of a bespoke model for classifying industrial components, integrated within an autonomous disassembly system. The model development process leveraged the available tools on the Roboflow online platform, such as accelerated annotation using SAM3, augmentation, and training. The model performed robustly in real-time, implemented using an eye-in-hand perception configuration, enabling accurate execution of robotic disassembly tasks.
AudioFine-tuningA short design project involving development of a CNN for audio classification. Real-life audio samples carrying information indicating medical conditions were provided, and the work package involved both data pre-processing and model architecture selection. Spectrogram analysis was incorporated, and a CNN with a combination of convolutional, pooling, and dense layers was used to enable robust performance.
ImageSegmentationA group project with five 5th-year undergraduate students, involving the implementation of a proof-of-concept autonomous refurbishment system. The system included an AI-driven perception pipeline to execute real-time inference from a custom-trained object detection model. Visual data of industrial components such as screw fasteners, panels, and gearbelts was collected, annotated using the accelerated tools on the Roboflow online platform (such as SAM3), trained, and implemented. Annotation combined manual and autonomous approaches to generate segment masks for all target classes, as well as forked datasets which were repurposed to meet the specifications of the designed model. Robust performance was verified through laboratory-based testing, using metrics of recall, precision and F1-score.
Master of Engineering, Electrical and Mechanical Engineering
Robotics Engineer Research Intern
Freelance Gardener
