Project Report (Master of Science/Engineering in Computer Science and Engineering)

Permanent URI for this collectionhttps://hdl.handle.net/10361/28443

Browse

Now showing 1 - 20 of 24

A practical approach to convert silo web application into multi-tenant cloud-based SaaS application
(BRAC University, 2025-10) Khan, Omar Hassan; Chakrabarty, Amitabha; Department of Computer Science and Engineering
Generally, SME software firms start developing an application with specific require- ments and intended to use it as Application Service Provider (ASP). With no argu- ment ASP fits best for agile development and meets development deadlines. In ASP model, we deploy the system and cater update, customization and troubleshooting by ownership. When we have the application deployed for more than a handful of users then it gets difficult to maintain the application for each owner. In Software as a Service (SaaS) model service is provided to each subscriber individually but under- neath the application is maintained as one single instance. Converting ASP model applications to SaaS model application helps the provider to reduce development and operational costs, effortlessly update and do bug fixes, and attain troubleshooting effectively. The conversion of ASP to SaaS is not straightforward. The conver- sion starts with separation common modules and subscriber specific modules. Then there is existing database customization along with master table implementation for the subscription. Finally, we need to select proper cloud-based IaaS with different services to obtain maximum usage with minimal billing. This project address the challenges of converting an ASP to a multi-tenant SaaS model with scalability, cost efficiency and ease of maintenance. Using the cloud as infrastructure gives the advantage of high security, reduces overhead cost and streamlines integration and deployment. This paper aims to provide a structured implementation technique to convert an existing singleton web application to SaaS on multi-tenant shared infrastructure model. These improvements help software firms achieve better operational efficiency while offering a more sustainable SaaS solution. Also helpful for the researchers and engineers to easily implement multi-tenant architecture in action.
Bangladesh travel bot: RAG application using LLM
(BRAC University, 2025-10) Siddika, Ayesha; Mostakim, Moin; Department of Computer Science and Engineering
The ”Bangla Travel-Bot” project aims to create an intelligent and user-friendly chatbot that provides accurate, context-specific responses to travel-related queries for Bangla-speaking travelers. As travel information in Bangla is often limited, this project seeks to fill the gap by offering a platform that ensures seamless access to vital travel information in the user’s native language. The chatbot is designed to assist travelers with a variety of needs, such as recommendations for tourist destinations, transportation options, accommodation, local customs, and real-time updates on weather, events, and travel logistics. This system leverages the power of a Retrieval-Augmented Generation (RAG) framework, which enhances the bot’s ability to generate high-quality responses by retrieving relevant information from trusted travel sources. The integration of LLaMA 3, a state-of-the-art language model, ensures the chatbot can effectively process and respond in Bangla, maintaining both linguistic accuracy and cultural relevance. The chatbot is capable of understanding user queries and providing personalized recommendations based on location, preferences, and travel context. Through the use of advanced natural language processing (NLP) techniques, the ”Bangla Travel-Bot” offers an intuitive conversational interface, ensuring that users can easily interact with the system. Whether it’s helping with last-minute travel plans or offering insightful tips about a destination, the chatbot strives to enhance the overall travel experience for Bangla-speaking users. This project not only aims to simplify the process of planning and managing travel for Bangla speakers but also contributes to the development of accessible, language-specific travel assistance tools in the broader field of artificial intelligence and human-computer interaction..
Benchmarking and enhancing Bengali OCR: a hybrid OCR system with analytic hierarchy process-based evaluation
(BRAC University, 2025-10) Pathan, Md. Keum Uddin; Alam, Md. Golam Robiul; Department of Computer Science and Engineering
This study benchmarks the performance of three OCR systems—Tesseract OCR, EasyOCR, and a hybrid approach combining Tesseract OCR, EasyOCR, and the Google Vision API—for Bengali text recognition. The evaluation was conducted on a diverse, real-world dataset comprising 216 images across nine categories of Bengali documents, totaling 19,064 words. Each OCR engine was independently assessed using multiple performance metrics, including Character Error Rate (CER), Word Error Rate (WER), Character-Level Accuracy (CLA), Word-Level Accuracy (WLA), and processing time. Among other preprocessing techniques, the pipeline employed grayscale conversion, resizing, noise removal, and adaptive thresholding; however, these steps did not consistently enhance the performance of standalone OCR engines. To address the limitations of single-engine systems, a hybrid OCR framework was developed that processes raw images and employs a multi-criteria decision-making approach based on the Analytic Hierarchy Process (AHP). A user study involving 41 participants was conducted to determine the relative importance of CER versus WER. Using Saaty’s scale, over 70% of participants assigned a value of 5 or higher in favor of CER. This resulted in a CER-to-WER importance ratio of 4.76, which was then used to compute AHP-based weights. For each image, OCR outputs were scored using a weighted combination of CER and WER, and the engine with the lowest score was selected as the optimal result. The hybrid system demonstrated strong performance under optimal conditions, achieving a Character- Level Accuracy (CLA) of 96.63% and a Word-Level Accuracy (WLA) of 80.34%, corresponding to a Character Error Rate (CER) of 3.37% and a Word Error Rate (WER) of 19.66%. This significantly outperformed Tesseract OCR (CLA: 88.54%, CER: 11.46%; WLA: 79.99%, WER: 20.01%) and EasyOCR (CLA: 90.98%, CER: 9.02%; WLA: 78.06%, WER: 21.94%). These results were obtained from specific document categories where OCR performance tends to be highest. While recognition accuracy may vary across different document types, the findings highlight the potential of the AHP-guided hybrid approach to substantially improve Bengali OCR performance in favorable scenarios and provide a strong foundation for further enhancement in more challenging, real-world conditions.
Voice-controlled browser extension using machine learning for enhanced accessibility
(BRAC University, 2025-05) Akter, Rabeya; Mostakim, Moin; Department of Computer Science and Engineering
This project presents a voice-controlled browser extension designed to enhance web accessibility and convenience through natural language voice commands. Targeting users with motor and visual impairments, as well as those seeking hands-free multitasking, the system integrates the Web Speech API for real-time speech recognition and TensorFlow.js for machine learning-based command interpretation, complemented by browser automation techniques. Its user-centered design incorporates intuitive voice commands and feedback mechanisms, ensuring the extension is approachable for non-technical users. The modular and scalable architecture facilitates easy updates and supports potential expansions, such as broader command sets, multi-language capabilities, and additional accessibility features like text summarization or translation. Key contributions include improved accessibility for users with disabilities, seamless support for multitasking, and the practical integration of interdisciplinary technologies. By successfully executing a range of browser commands, this work advances human-computer interaction and underscores the transformative potential of voice-driven interfaces in creating more inclusive digital environments.
Design and development of an intelligent loan eligibility prediction system using machine learning & explainable AI
(BRAC University, 2025-02) Dey, Arnob Kumar; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
It can be seen that the value of assets is rising daily. That often necessary a lot of capital to buy the whole asset. Many times it becomes impossible to buy even from our savings. So we can apply for a loan to get the money we need because, through the loan application, we can easily get the money for buying assets. However, obtaining a loan is a lengthy procedure. The application must go through several steps before being accepted, and approval is not guaranteed. Many loan prediction models have been created to reduce the time and also reduce the risk attached to the loan. The main goal of my project was to compare different types of prediction models and then finalize which one is the best for loan-eligible prediction with the lowest amount of mistakes. Also used SHAP method for Feature importance of model prediction. After that choose the best ML model for deployment. So, I used eight machine learning models for my research paper. I get Logistic Regression is the best among the eight machine learning models with high accuracy and F1 score. For my application, I used this logistic regression model for prediction and also used SHAP for feature importance. After that, used text generation model L Lama-3.3-70b-versatile with Groq API to explain the reason for the prediction and give suggestions based on the prediction.
AI thesis helper
(BRAC University, 2024-12) Shayer, Mirza Ahmad; Mostakim, Moin; Department of Computer Science and Engineering
In the modern world, Artificial Intelligence (AI) has become popular. Machine Learning (ML), is another popular term used on the internet, it is when AI is applied to a system which allows it to learn. Deep Learning (DL) is another popular term that became very common in recent times, it is the application of ML that utilizes algorithms and trains models. In 2020, Generative Pre-trained Transformer (GPT), came to the mainstream and had become highly successful. In just another year it became a game changing tool in the scientific landscape. Using GPT, various chatbots are designed. Chat-GPT in particular is by far the most used and most useful model of GPT. The model is useful in generating replies to various questions with high accuracy and usefulness. Can be used to help users with almost practically any question. In a few years this model will develop to help people even further. As such, this project - AI Thesis Helper, is one such implementation of further helping students to aid their thesis writing. Many students struggle to write their thesis, as such this chat-bot can help students with various thesis tasks like idea generation. This project aims to assist students in their writing. Project challenges and objectives are elaborated. Prior works are looked into along with reasons for choosing GPT over other models. Requirements, pros, cons and feasibility - both economic and technical along with model diagrams are dealt with. Details on the dataset are discussed. Fine-tuning of the model is elaborated with all details like imports, hyperparameters and training loop etc. Front-end, integration and backend are discussed along with UI/UX components. Implementation shows the project workings and how well it performs. To evaluate the performance, the following metrics are utilized - model loss during training, precision, recall, F1 score, perplexity and spelling, grammar with overall writing quality. Along with a comparison of generated text, made with two other AI chatbots - ChatGPT and Perplexity AI for the same text input. Future updates are also described.
Parameter-efficient fine-tuning of LLMa-2 using quantization low-rank adaptation
(BRAC University, 2024-11) Islam, Md. Tariqul; Mostakim, Moin; Department of Computer Science and Engineering
Llama-2, an advanced neural network with huge potential in text generation, sentiment analysis and language understanding. This report focuses on the fine-tuning process for build chatbot on custom datasets, specification methods, hyperparameters and training strategies. Experimental results on Guanchu datasets show excellent adaptability of the model, outperforming the baseline model in human evaluation and achieving significant BERT scores for help and safety. The analysis includes an in-depth examination of LAMA-2’s architecture, outlining strengths and suggesting areas for improvement. Parameter-efficient fine-tuning and quantization also investigate the transformative potential of LLMA-2 through low-rank adaptation. The objective is to strike a balance between model complexity and efficiency, addressing challenges in resource-constrained environments.
Ahar24: a modern restaurant food delivery management system
(BRAC University, 2024-11) Iqbal, Syed Bayes; Mostakim, Moin; Department of Computer Science and Engineering
The rapid digital transformation has significantly impacted the food and restaurant industry, leading to the emergence of online food delivery systems. This research presents the development of Ahar24: A Modern Restaurant Food Delivery Management System, a comprehensive platform designed to streamline food delivery operations and enhance user experience. The system integrates advanced features such as personalized food recommendations powered by AI, a smart menu search with natural language understanding, dynamic pricing and discounts, and real-time order prediction and optimization. These features cater to both customers and administrators, ensuring seamless interactions, efficient order management, and data-driven decision-making. For customers, the platform offers a user-friendly interface for browsing menus, placing orders, and accessing tailored food suggestions based on preferences and past behavior. Administrators benefit from robust management tools, including insights into customer behavior, trending items, and demand forecasting. The system is implemented using cutting-edge technologies such as React.js, Node.js, and MongoDB, ensuring scalability and efficiency. While still under development and testing, Ahar24 demonstrates the potential to redefine modern restaurant operations and food delivery services by incorporating innovation, user-centric design, and advanced analytics.
Online music application with modified features of Bangla Songs
(BRAC University, 2024-05) Khandaker, Moumita; Hossain, Muhammad Iqbal; Department of Computer Science and Engineering
The musical industry has been transformed through a revolutionary change over the past few years due to the increasing availability of the internet across the globe. In this phenomena, musical streaming mobile application plays a significant role. It is easy to use, hassle free to download, effortless streaming and sharing experience and gives the musical industry a new platform of promotion and earning revenue along with the user benefits. Some of the most famous Mobile Applications used all around the world are Spotify, Apple Music, Sound Cloud etc. The common features in all these applications are huge libraries of music of various countries, genres, artists, customizable playlists, and machine learning algorithms applied in understanding each user’s preferences for song recommendations. These Applications not only foster the listeners’ benefit but also provides a revenue platform to the artists. They launch services like subscription selling, advertisement generation, promotional strategies, patron support from listeners etc. that expand business growth. Although there are issues faced very commonly such as artist compensation, user privacy maintenance, streaming issues etc. My proposed Music Application has provided a streaming platform to not only surf the international songs but also an individual module to browse for Bangla songs. There is unavailability of Bangladeshi Music Mobile Application that is a hybrid platform for both local and foreign artistry. This application provides some unique features along with an efficient business model that makes it stand out on the market and holds potential for a futuristic approach.
PlantGuard: intelligent plant disease detection
(Brac University, 2024-05) Khanom, Nazifa; Hossain, Muhammad Iqbal; Department of Computer Science and Engineering
Every year, there is significant crop loss in developing countries due to delays in identifying plant diseases. Prompt and accurate identification of these diseases, with less reliance on field experts, could greatly mitigate this issue. Recognizing plant diseases correctly, particularly when they present similar leaf textures, poses a significant challenge. It’s crucial to consider factors such as leaf color and various texture features to accurately predict plant defects. The objective of this project is to employ Deep Learning methodologies for the detection of plant diseases based on leaf images. Deep learning, specifically Convolutional Neural Networks, is chosen due to its effectiveness in extracting features from plant leaves, making it well-suited for image data analysis in this context.
Full stack development of UBERA system
(BRAC University, 2023-05) An Noor, Akib; Alam, Md. Ashraful; Department of Computer Science and Engineering
The system named ‘UBERA’ is a web application which is undertaken as a system which has the ability to assess buildings and produce an output result if they are vulnerable to earthquake or not. It is an automatic decision making system that evaluates the risk of buildings during an Earthquake. It utilizes modern Artificial Intelligence techniques along with civil engineering concepts to make the decision. A detailed structural evaluation is recommended to any building flagged as by the system. For building evaluations an effective system should be introduced in order to address this alarming earthquake risk. And with the help of this system anyone can access the system to know about any information about any types of buildings, its structural condition, awareness related information, evaluation facilities, automatic image processing result etc. This web application has several features, from google map API integration to the image processing facility of getting proper building evaluation results through the system. Moreover, generating a complete report page, maintaining the FEMA standard to calculate the threshold automatically are being developed as features of the system to make it an all in one building assessment and evaluation platform. The aim of the application is to develop awareness amongst general people about their building for any structural defects that might make the building vulnerable to earthquake and provide information with a proper report about the total building conditions.
An android application to predict human activity using a deep learning LSTM model
(Brac University, 2023) Sikder, Debabrata; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The machine learning approach to estimate human activity using smartphone sensor data is challenging. In this project, the HAR approach is conducted based on the LSTM model and can recognize six different behaviors, i.e., Downstairs, Jogging, Sitting, Standing, Upstairs, and Walking. To achieve the best potential result, various machine learning and statistical approaches were explored. The long shortterm memory (LSTM) is a recurrent neural networks (RNNs) capable of learning long-term dependencies, especially in sequence prediction problems. This LSTM model was applied in this project, to obtain the desired result. This model shows 97% test accuracy. Finally, the model was exported and deployed in the Android application, which has an user interface that could provide a user-friendly experience.
Resort management system
(Brac University, 9/9/2022) Hasan, Faysal Bin; Hossain, Muhammad Iqbal; Department of Computer Science and Engineering
Electronic voting system based on blockchain
(Brac University, 2022-09) Ahmed, Tanvir; Mostakim, Moin; Department of Computer Science and Engineering
The primary goal of this application is to create a voting app which is safe and the result is unchangeable. By using blockchain technology these goals can be achieved. Data in blockchain is not stored in only one location and the administration responsibility do not belong to any person. All the data of a blockchain is stored in different locations. The blockchain is instead duplicated and dispersed among many computers in a network. Every computer on the network updates its blockchain each time a new block is added to the blockchain to reflect the change. In this application anyone can create a voting poll and the poll will be added to a block. People visiting the app will be able to see the polls and can cast a vote to the favorite candidate. And people will get to choose a winner based on the amount of vote on candidate got. Because this application is created using blockchain technology, everyone can trust this application and will be satisfied with the result they get as the result will be unbiased. This application will be useful to anyone who is willing to shere their opinion and is willing to compare popularity of different persons.
Skin lesion classification using different CNN models
(Brac University, 2023-05) Oli, Md. Yahea Sultan; Chakrabarty, Amitabha; Department of Computer Science and Engineering
The field of dermatoscopic image classification has gained significant attention as there is a growing demand for early diagnosis of specific diseases. The use of deep learning is increasingly significant in the quest for a more effective dermoscopic anal- ysis method. The “HAM10000” (Human Against Machine) dataset has been used in this study for classification of 7 different types of skin lesions by using DenseNet-121, VGG16, ResNet50, and Inceptionv3 model. To improve the classifier’s performance, data augmentation was applied. This study could help dermatologists in the clinic make more precise decisions when identifying skin lesions, which would be benefi- cial. With this project I have tried to improve the model so that dermatologists identify skin lesions more precisely. Through the implementation of data augmen- tation techniques, this project achieved an impressive categorical accuracy of 92% and a top2 accuracy of 97% using DenseNet-121. The remaining models, VGG16, ResNet50, Inceptionv3 achieved accuracy 80%, 78%, 84% respectively. This project could have a beneficial impact on dermatoscopic image recognition and can reduce time and valuable resources. It can also help to saves life where robust diagnosing is not available.
References validation in scholarly articles using RoBERTa
(Brac University, 2023-05) Nasib, Abdullah Umar; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
In the recent era of technological advancement, evaluating references and assignments and validating those are some of the primary processes to justify the authenticity of a research work in the academic sector. The purpose of referencing is to ensure ethical writing and to make the search easier in a particular area for the reader when it is accurate. The traditional way to cross-check the validations is to check them manually one by one, which is above argument, equivalent to an- other work, and sometimes so tiring that the reader loses interest in reading the original paper. The target of this research paper is to introduce a semi-automatic digital system that enables researchers to justify the references used in a research paper without doing it manually. In this model, we have prepared a sentence trans- former named RoBERTa for generating embedding. We sanitize and preprocess an entire research paper and cross-match that against a reference query using the proposed model in terms of finding semantic and contextual similarity. The result shows mostly similar contexts based on the similarity check. We have compared the embedding of query and user input articles with the help of K similar search function. Our model outperformed the existing BERT and SBERT models’ output in accuracy with a F1 score of 0.777, which establishes the fact that the model can be used in real life with a simple query of text from research articles.
Effective feature selection for real-time stock trading in variable time-frames and multi criteria decision theory based Efficient stock portfolio management
(Brac University, 2021-08) Ullah, A. K. M. Amanat; Alam, Md. Golam Rabiul; Department of Computer Science and Engineering
The unpredictability and volatility of the stock market render it challenging to make a substantial pro t using any generalized scheme. Many previous studies tried di erent techniques to build a machine learning model, which can make a signi cant pro t in the US stock market by performing live trading. However, very few studies have focused on the importance of nding the best features for a particular period for trading. Our top approach used the performance to narrow down the features from a total of 148 to about 30. Furthermore, the top 25 features were dynamically selected before each time training our machine learning model. It uses ensemble learning with four classi ers: Gaussian Naive Bayes, Decision Tree, Logistic Regression with L1 regularization and Stochastic Gradient Descent, to decide whether to go long or short on a particular stock. Our best model performed daily trade between July 2011 and January 2019, generating 54.35% pro t. We further propose a novel model which uses Ada-boost to nd the weights of each of the features and then apply TOPSIS to select the best stocks. Lastly, we survey the machine learning techniques used for ethical decision-making in stock trading, which will bene t any further research work on Responsible AI in Finance.
A framework of biometric recognition and personalized mobile application to establish a dynamic connection with the ATM to enable secure transaction
(Brac University, 2019-08) Tasnim, Syeda Prima; Rahman, Md. Khalilur; Department of Computer Science and Engineering
Secure transaction of money is the prime concern of banking industries for establishing a strong banking network worldwide. By the vast uses of Automated Teller Machine (ATM) in this industry transaction of money has become easier for the customers. But the system needs more secure transaction method for avoiding unwanted incidents. Smart and secure transaction through ATM is the thing people want now a days. For making the transaction through ATM more reliable as well as smart a biometric based system along with an android application has been approached in this research paper. The system consists of a fingerprint-based device which is connected with the ATM software and a personalized mobile application for user. Using the application customer can make a 1:1 connection with the ATM software. For transaction customer can control the account using the mobile application and make transaction command which is also connected with the central server. At the very last stage customer needs to give finger print on the designated device for verifying the account holder’s presence during the transaction. In this way the less safe ATM card system can be replaced with a smart and secure system. Testing of the system using a custom-made ATM software along with fingerprint sensor justify the authenticity as well as efficiency of the system.
Online shopping management system
(Brac University, 12/24/2017) Sultana, Akhi; Uddin, Jia; Department of Computer Science and Engineering
The primary goal of this e-commerce site is to sell goods and services online. This project deals with developing an e-commerce / Online Shopping Management System for online product sale. It will provides the user with a category of different products available for purchase in the store. ln order to facilitate online purchase a shopping cart will provided to the user. The system is implemented using a 3-tier approach, with a backend relational databases (such as MySQL), a middle tier of Object-Oriented PHP, and a web browser as the front end client. This is a project with the objective to develop a basic website where a customer is provided with a shopping cart application and also to know about the technologies used to develop such an application. This system involves its own database to be maintained. The information or details about the products are stored in the database. The server process is for dealing with the customers detail and the items that are shipped to different locations based on the addresses provided by the customers. The application design contains two modules one is for the customers who wish to buy the product and another is for the store owners who maintains and updates the information regarding to the product and about the customers. The end user to use this site are the common people for whom this application is to be hosted on the web and the admin maintains the database.
Nutrition system – a mobile application for pregnant woman
(BRAC University, 2018-05) Anam, Md Toriqul; Chakrabarty, Amitabha; Department of Computer Science and Engineering
Malnutrition has a devastating impact on the health and wellbeing of individuals and on the economic development of countries. In Bangladesh there is scarcity of proper nutrition information system that can provide appropriate nutrition messages based on different criteria of pregnant woman and infant while a primary or tertiary health workers could not provide the necessary service to them. This application can be easily used by any health workers in Bangladesh. Our health workers often forget to deliver the proper nutrition information to mothers. Such tool can be very useful to provide a proper way to deliver the specific nutrition messages to mothers based on their Ante-natal care (ANC), Post-natal care (PNC) state and also based on their infant age. The model of this application can provide the proper facilities of delivering nutrition messages to mothers by health workers. This system may need to analyse periodically in order to meet the changes of user requirement and apply appropriately.

Browse

Recent Submissions