Kazi M.

Key Skills

Software

No software listed

Top Subject Matter

Human action classification / video understanding

Humanoid-robot AI / human-robot interaction

Top Data Types

Video

3D Sensor

Image

Top Task Types

Action Recognition

Data Collection

Classification

Freelancer Overview

MSc Researcher — Human Action Classification (University of South Wales). Brings 5+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include OpenCV, PyTorch, and Flask. Education includes Master of Science, University of South Wales (2023) and Bachelor of Science, Daffodil International University (2019). AI-training focus includes data types such as Video and 3D Sensor and labeling workflows including Action Recognition, Data Collection, and Classification.

Labeling Experience

MSc Researcher - Human Action Classification - University of South Wales

VideoClassification

Designed an end-to-end video-action recognition workflow including frame sampling, resizing, temporal handling, and training-validation splitting. Trained and evaluated 3D CNN models using UCF101 and Kinetics-700 datasets, analyzing results with confusion-matrix evaluation. Deployed the trained model through Flask APIs and connected it to web and Android interfaces for practical video upload and prediction. • Built video preprocessing and training pipeline • Trained and evaluated 3D CNN action classification models • Performed model analysis using confusion-matrix evaluation • Integrated and deployed via Flask, web, and Android clients

2023 - 2023

MSc Researcher — Human Action Classification (University of South Wales)

VideoAction Recognition

Designed and implemented an end-to-end video-action classification workflow for training and evaluation on human action datasets. The work included preparing video inputs, managing temporal dimensions, and analyzing model performance to support action recognition tasks. Outputs were deployed via Flask API and client interfaces for practical video upload and prediction. • Framed preprocessing steps including frame sampling, resizing, and training-validation splitting • Trained and evaluated 3D CNN models on UCF101 and Kinetics-700 video datasets • Performed confusion-matrix analysis to assess recognition quality across action classes • Integrated model inference into Flask API, web UI, and Android client workflows

2023

Robotics Lab Assistant (Daffodil Computers Ltd)

3D SensorData Collection

Collected and organized training data to support humanoid-robot AI experiments and human-robot interaction research. The preparation focused on making collected data usable for downstream model testing and response evaluation. Work involved ensuring data cleanliness and structuring for reliable AI experimentation. • Collected and cleaned training data for humanoid-robot AI and HRI experiments • Organized datasets to support consistent training and evaluation pipelines • Built Python/Java components to support response testing and input/output handling • Implemented automation and real-time communication routines to facilitate iterative data-driven testing

2019

Education

Work History