Kazi Mahathir Rahman

In our project, we tackled the challenge of detecting ships from Synthetic Aperture Radar (SAR) images using the YOLOv7 model. SAR images are often plagued by significant noise. To mitigate this, we applied noise reduction techniques such as Denoising Diffusion Probabilistic Models (DDPM) and anisotropic diffusion, which significantly enhanced image quality. Our model was trained on SSDD, DS SDD, and FUSAR datasets and demonstrated high accuracy in ship detection.

This project investigates facial expression recognition using Swin Transformer models (Swin-S, Swin-M, Swin-B). The system initially employs Haar Cascade for face detection, followed by Swin Transformer-based classification on datasets like RAF-DB, CK+, and FER2013.

PyParallel is a PyTorch-based library designed to simplify deep learning on multiple devices with features for model and data parallelism, allowing efficient training across CPUs, GPUs, and TPUs. It includes optimization tools like gradient checkpointing, mixed precision, automatic loss scaling, gradient clipping, and gradient accumulation to improve performance and memory usage. Additionally, PyParallel offers real-time hardware monitoring, providing CPU, GPU, and TPU memory tracking through an interactive dashboard, making it easy to manage resources during training.

This project aims to predict player positions in FIFA 23 by employing various data preprocessing techniques such as normalization, scaling, feature engineering, and handling missing values on a Kaggle dataset. Several machine learning models including Random Forest, Support Vector Machines, and Neural Networks were compared to optimize player position prediction accuracy.

BRACU Helper is a university chatbot built to assist BRAC University students by answering their common queries. Developed using Django for the backend and React for the frontend, the chatbot utilizes a fine-tuned Sentence Transformer model to generate accurate and relevant responses. The project includes features like user authentication, a responsive chat interface, and data storage for chat logs, making it a useful tool for enhancing the student experience.

EssenceAI is a text summarizer website developed using Django and React, designed to help users generate concise summaries of large bodies of text. The site leverages a custom-trained T5 model, fine-tuned with the Gigaword dataset, ensuring high-quality and contextually relevant summaries. The interface is built using the Bootstrap CSS framework, offering a modern and responsive design.

This project is an Islamic Knowledge Assistance Chatbot built using Django, HTML, and CSS, designed to provide accurate answers to questions related to Islam. The chatbot employs a retrieval-based approach utilizing TF-IDF vectorization and cosine similarity, along with natural language processing techniques like tokenization and stopword removal, to match user queries with the most relevant answers from a custom dataset of Islamic questions and answers.

This project focuses on sentiment analysis of a dataset using both machine learning and deep learning models, including Random Forest, RNN, LSTM, and GRU. Despite employing techniques like dropout, regularization, and hyperparameter tuning, all models exhibited overfitting. Improvements in data preprocessing and acquiring a more diverse dataset are necessary generalization to unseen data.

This project aimed to investigate the relationship between TripAdvisor reviews and corresponding ratings. By employing a combination of Natural Language Processing (NLP) techniques and deep learning models, the study sought to understand the extent to which review text could predict or explain rating variations.

This project is a web scraping bot designed to extract product details from the Daraz online shopping platform. By providing a search keyword and specifying the page number, the bot uses Selenium to navigate Daraz's website, collect data such as product name, brand, price, and seller information, and then saves this data into a CSV file.

This project involved web scraping mobile phone data from Bikroy.com using Python. Key information such as mobile name, model, price, reviews, features, and other details were extracted from the first five pages of the website due to hardware constraints. The Python libraries Beautiful Soup 4, requests and selenium were primarily used for this task.

This project explores the development of object detection models using ResNet-34, ApexNet, InceptionNet, and GoogleNet architectures build from scratch on a specified image dataset utilizing TensorFlow.

This project investigated the potential of machine learning for breast cancer analysis and prediction. By exploring various models like SVM, Logistic Regression, KNN, and Random Forest on a preprocessed dataset, the project aimed to identify the most effective approach. While all models achieved decent accuracy, SVM stood out with a superior F1 score, indicating its better ability to balance precision and recall.

This research investigates the prediction and classification of protein sequences using a dataset encompassing diverse protein information. A comprehensive approach was employed, incorporating various preprocessing techniques to prepare the data effectively. Multiple machine learning models were then applied and compared to determine the most suitable method for accurately predicting and categorizing protein sequences based on their underlying characteristics.

This project explores the potential of machine learning for stock price prediction using the S&P 300 index as a dataset. Random Forest, Logistic Regression, and Support Vector Machine models were implemented and compared to forecast future stock prices. The project encompasses data collection, preprocessing, feature engineering, model training, evaluation, and visualization.

This platform lets users track fitness goals and book household services like cleaning, repairs, and gardening. Built with PHP, MySQL, HTML, and CSS, it offers easy management for health and home tasks in one place.

I've constructed a dynamic online portfolio showcasing my web development abilities. Leveraging HTML, CSS, JavaScript, and PHP, I've crafted a responsive and interactive platform hosted on GitHub. This digital showcase effectively highlights my skills and projects, providing a comprehensive overview of my professional capabilities.