• ◈ Developing and managing end-to-end Machine Learning systems, focusing on designing solutions, collecting data and fine-tuning LLMs, VLMs and Computer Vision models for enhanced availability of test automation.
• ◈ Contribute to Patent and ARF Projects, promoting an innovative culture aligned with the company’s core values.
• ◈ Coordinated and instructed internal Generative AI training, leading a group of 30 employees.
• ◈ Collaborating with a supportive team to foster growth and enhance engineering capabilities
• ◈ Simultaneously managed two projects, contributed to two patents, and created multiple proof-of-concept tools.
• ◈ Recieved Excellence Award for Innovation from Samsung R&D Institute

• ◈ Analyzed, designed and developed an "Automated Test Case Generation and Execution Tool" for Android mobile applications.
• ◈ Designed a Custom Depth First Search algorithm to efficiently traverse an application, write test steps in the process and find UI bugs.
• ◈ Determined the lack of performance in using Appium in our algorithm and designed solutions for performance optimization.
• ◈ Fine-tuned multiple Machine Learning models for understanding screen representations and understanding unknown elements. Worked with MaxVit, EfficientNet and Flan-t5 in the process.
• ◈ Designed data collection protocols, trained and supervised data collectors, and ensured data accuracy and completeness.

• ◈ Implemented new features and customised front-end of online payment portal
• ◈ Analysed codebase for performance optimization

This research project, conducted at Samsung R&D Institute, focused on evaluating the effectiveness of large language models (LLMs) in executing test cases. We developed a framework that includes a robust approach for identifying the relevant UI elements needed to perform actions based on test cases using LLM, followed by interacting with these elements using Selenium. Subsequently, we applied semantic similarity-based textual matching between the final screen resulting from the UI interactions and the expected outcome defined in the test case to generate a verdict. My specific contributions involved simplifying web pages into a more generalized format for LLM comprehension, conducting few-shot prompting experiments with LLMs (such as, Solar-10.7B, StableBeluga-13B, LLaMa-2-7B), preparing dataset, fine-tuning the LLMs (LLaMa-3-8B, Solar-10.7B) model on annotated dataset for UI element identification and generating semantic similarity-based verdicts using all-MiniLM-L6-v2 (Sentence Transformer model), given the final screen and expected result. Our approach successfully executed 106 out of 110 targeted test cases, achieving an accuracy of 97%.

[White Paper]

This research project, conducted at Samsung R&D Institute, investigates the efficacy of Large Language Models (LLMs), Vision-Language Models (VLMs), and Computer Vision Models in generating test cases from UI guides. We developed a framework that employs a robust method for detecting functional flows within software from UI guides using Vision-Language Models (VLMs), Large Language Models (LLMs), and Computer Vision models, converting these functional flows into test cases through an algorithmic approach. Additionally, we generated expected outcomes for specific functional flows using VLMs. My contributions included conducting few-shot prompting experiments and fine-tuning VLMs (such as CogVLM and Idefics2) for UI screen identification and functional flow mapping. I also performed UI screen matching with their corresponding descriptions using the LLaMa-3-8B model. Our framework effectively generates test cases from extensive UI guides with minimal error and within a reasonable time frame.

This project is a collaboration between Samsung R&D Institute and BUET where we studied the effectiveness of Large Language models in generating and executing test cases for android applications under the supervision of Dr. Anindya Iqbal. We developed an automated UI testing tool for Android applications that can significantly reduce manual testing efforts by 80% by automatically generating test cases. The tool utilizes a custom DFS algorithm using Appium to traverse the application's UI and write test steps. Additionally, EfficientNet and MaxViT are employed for image classification of unknown screen elements and screen similarity matching for maintaining DFS state, while the Flan-T5 generative model is used to understand screen representations and generate expected results for each test step.

[White Paper]

Undergraduate thesis project under the supervision of Dr. Anindya Iqbal and Dr. Shahrear Iqbal (Research Officer, National Research Council (NRC) Canada). In this research, we created a framework to efficiently process sequences of system calls, making them more accessible for language models to identify malicious patterns and detect Android malware. We utilized a Transformer models to analyze these sequences for malware detection. We fine-tuned various language models (BERT, RoBERTa, BigBird, LongFormer) with different sequence lengths and techniques such as supervised fine-tuning and contrastive learning. We then compared their performance against existing models like LSTM, Random Forest, and SVM. Our evaluation demonstrated a 6-7% improvement over the previous state-of-the-art model.

[Thesis dissertation]

This is a project under the supervision of Dr. Anindya Iqbal where our aim is to produce more human like reviews and automating review generation procedure through large language models. We are working on the latest RL based preference optimization to finetune state of the art open source LLMs and figure out more suitable evaluation metric to find out human likeliness of code review. The knowledge domains we need in this study are: Code review, LLM inference, LLM finetuning and Objective evaluation.

CGPA: 3.75/4.00

GPA: 5.00/5.00

GPA : 5.00/5.00

I organized internal Generative AI training at Samsung R&D Institute with 30 selected employees to enhance their skills in the Generative AI domain. My role included coordinating lecture assignments with other trainers and leading individual training sessions. Notable sessions I conducted included topics like Introduction to Deep Learning and ANN, Image Processing with CNN, Introduction to LLM, Prompt Engineering, and Fine-tuning LLMs. Our efforts were well-received, earning recognition from both the Managing Director of Samsung R&D Institute, Bangladesh and Korean Headquarters.

Organized the first ever deep learning competition to be hosted in BUET CSE Festivals. My responsibilities ranged from setting up environment to collect dataset for the competition to training and supervising data collectors and ensuring data accuracy and completeness. We partnered with Bengali.AI and hosted this competiton in Kaggle. The competition was a resounding success with 122 competitors comprising of 59 teams and 470 submissions. Kaggle recognized our efforts by providing us the Kaggle Community Competition Creator Prize

Played a key role in organizing the Science Fair 2016 at Dhaka Residential Model College.

Tutored many students ranging from middle school to high school and undergrad-level students during my undergraduate years.