I've dedicated my efforts to designing software aimed at enhancing the quality of life. Through
active participation in various software development groups, I've had the opportunity to
contribute to several impactful projects. Here are a few examples of my designs across different
programming languages:
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
It is aptly named for its purpose within the water purification industry. This software serves the crucial role of managing an inventory of water purification units. Primarily utilized in warehouses where employers seek efficient tracking of their products, it ensures seamless organization and accessibility of inventory data. *******************************************************************************************************************************************
I developed this software using the Spring Boot framework, which efficiently manages the core database operations. The application boasts an aesthetically pleasing interface, enhancing the overall user experience on the website. It facilitates the addition of student records to an online PostgreSQL database hosted on the Rander website, incorporating specific attributes such as name, weight, height, hair color, and GPA. The architecture of the software is solidly built on the Model-View-Controller (MVC) pattern, serving as the backbone of the code. It leverages the capabilities of Spring Data JPA repositories for data persistence and manipulation. Furthermore, it employs GetMapping and PostMapping annotations to handle HTTP GET and POST requests, respectively, enabling seamless database operations. *******************************************************************************************************************************************
It is aptly named for its purpose within the breast cancer detection field. The model accurately predicts the malignancy of tumors, achieving an impressive accuracy rate of over 95%. *******************************************************************************************************************************************
This project involves using machine learning to classify images measuring 28x28 pixels, where each image represents a single object. These images belong to various fashion categories, such as T-shirts, pullovers, bags, trousers, and other clothing or accessory types. The process typically involves training a neural network or similar algorithm on labeled datasets to recognize patterns, shapes, and features unique to each category. Once trained, the model can accurately identify the type of object depicted in new, unseen images. This application is common in fashion-related AI projects, often using datasets like Fashion-MNIST for training and testing. *******************************************************************************************************************************************
This project simulates a vacuum cleaner agent that uses various AI algorithms to search and clean a room. It evaluates the time efficiency of these algorithms in performing the cleaning task. The code is written in Python and is available on my GitHub page. *******************************************************************************************************************************************
This neural network model is designed to predict insurance charges by analyzing various factors such as age, sex, BMI (Body Mass Index), and more. To ensure accurate predictions, the model uses normalized data, which scales the input features to a consistent range, improving the efficiency and performance of the learning process. By understanding the relationships between these factors, the model provides reliable estimates of insurance costs for individuals. *******************************************************************************************************************************************
This project is a TicTacToe game designed to create a highly intelligent AI agent. I implemented various algorithms, including Minimax, Alpha-Beta Pruning, and Monte Carlo Tree Search. As a result, the AI agent can now play the game efficiently on higher-order grids such as 4x4 or 5x5. The code is available on my GitHub account. *******************************************************************************************************************************************
The MNIST handwritten digit classification project is a foundational exercise in the field of machine learning, particularly valuable for those venturing into the realm of image recognition and deep learning. This project utilizes the MNIST dataset, which comprises a collection of 70,000 labeled images of handwritten digits from 0 to 9. Each image is a 28x28 pixel grayscale representation of a digit, making the dataset a manageable size for experimenting with various machine learning models, especially neural networks. A typical approach involves using a convolutional neural network (CNN), which excels in capturing spatial hierarchies in images. The objective is to train the model to accurately predict the digit represented in each image. This project not only serves as an excellent introduction to key concepts such as data preprocessing, model training, and performance evaluation but also provides a practical benchmark for comparing different algorithms in image classification tasks. *******************************************************************************************************************************************
Object Detection Using Classification: Object detection through classification involves identifying objects within an image by classifying them into predefined categories. In this approach, the image is divided into smaller regions or patches, and each region is classified as either containing an object of interest or being background. The classifier, typically a convolutional neural network (CNN), is trained to recognize various objects and can output the probability of each object class for every region. Non-maximum suppression (NMS) is often used to eliminate redundant detections, ensuring that only the most confident predictions are retained. This method is relatively straightforward but may struggle with detecting objects in varying sizes, positions, and scales, as the regions are typically fixed in size. Object Detection Using ResNet: ResNet (Residual Network) is a deep neural network architecture that has been highly successful in various computer vision tasks, including object detection. When applied to object detection, ResNet is typically used as the backbone of the network, meaning it serves as the feature extractor. The network is trained to identify and encode features from the input image into high-dimensional feature maps. These feature maps are then fed into a detection head (such as Faster R-CNN, YOLO, or SSD), which is responsible for predicting the bounding boxes and classes of objects within the image. ResNet's use of residual connections helps to mitigate the vanishing gradient problem, allowing the network to be deeper and more accurate. By learning residual functions instead of direct mappings, ResNet can efficiently capture complex patterns in images, making it particularly effective for object detection tasks that involve recognizing objects at multiple scales and in challenging conditions. This combination of ResNet for feature extraction and a detection head for object localization and classification results in a powerful and efficient object detection model. *******************************************************************************************************************************************
The face mask
detection system is
a computer vision application that utilizes machine learning to identify
whether individuals in video feeds or images are wearing face masks. This capability is
particularly
relevant in environments where health and safety regulations mandate the use of masks, such
as during
pandemics or in healthcare settings.
Key Components:
Data Collection:
A dataset is assembled containing images of individuals with and without face masks. This
dataset may
include variations in mask types, lighting conditions, and backgrounds to improve the
robustness of the
model.
Preprocessing:
Images are preprocessed for normalization, resizing, and augmentation (like rotation and
flipping) to
enhance model training effectiveness and to simulate diverse scenarios.
Model Training:
A convolutional neural network (CNN) is typically employed for this classification task. The
network
learns to extract features that are relevant for distinguishing between masked and unmasked
faces.
The training process involves dividing the dataset into training, validation, and test sets
to evaluate
the model's performance and prevent overfitting.
Classification:
The trained model classifies each face it detects as either "mask" or "no mask." Advanced
models might
also include classifications for improper mask usage (e.g., nose uncovered).
Deployment:
The final model is deployed into a real-time system where it can process live video feeds
from cameras,
providing alerts or access control based on mask compliance.
Use Cases:
Health Monitoring: In airports, hospitals, and other public spaces where mask enforcement is
critical.
Access Control: Integrating with security systems to allow entry only to individuals wearing
masks.
Analytical Insights: Gathering data on mask compliance rates over time or across different
locations.
This application combines image processing, deep learning, and real-time analysis to
contribute to public
health and safety measures effectively.
*******************************************************************************************************************************************
This project leverages PyTorch to train a deep learning model for breast cancer detection. The model is highly accurate, achieving an impressive accuracy rate of over 97%. By using PyTorch's robust framework, the project efficiently processes and analyzes medical data, providing reliable predictions to aid in early detection and diagnosis of breast cancer. *******************************************************************************************************************************************
This project is an advanced version of Multiclass Classification for Fashion Detection. While the original version likely used traditional neural networks or simpler models, this iteration employs a **Convolutional Neural Network (CNN)** for model generation. CNNs are particularly well-suited for image processing tasks, as they can efficiently capture spatial hierarchies and patterns, leading to improved performance and accuracy in classifying fashion categories such as T-shirts, bags, trousers, and more. *******************************************************************************************************************************************
Course planner (private repository) is an online course planning tool that utilizes actual course offering data from SFU. This software is developed using Spring Boot. It enables users to browse available courses, add new course offerings, and view graphs depicting the number of seats taken each semester. *******************************************************************************************************************************************
This is game (private repository) that allows multiple players shoot to the forts. A person who reaches to 2500 points wins the game. It is written in Java. My design is perfectly suited for this type of game. It is dynamic and innovative, tracking player interactions intelligently. I use a class named Opponent that contains all the attributes required by a game player. Additionally, it includes a hidden private variable that stores the coordinates of their forts. This design simplifies the process of checking each shot to determine if it's a hit. Moreover, my architecture includes a class named GameLogic which, as the name suggests, encapsulates all the logic for the game. This approach minimizes the amount of code needed for implementation, resulting in lightweight software that can run on all platforms with minimal specifications. Furthermore, my code makes extensive use of Streams, offering excellent practice for developers. *******************************************************************************************************************************************
Shape model is an innovative software tool designed to visualize and construct geometric shapes from specified input files. It meticulously reads each file line by line, interpreting the data to accurately render the corresponding shapes. The software supports a variety of file formats, ensuring compatibility and ease of use. It is particularly useful for designers and engineers who require precise graphical representations of shape models based on textual descriptions. Additionally, Shape Model features a user-friendly interface that simplifies the process of loading and modifying the shape files, enhancing productivity and creative exploration. *******************************************************************************************************************************************