Tasnia Sultana
Data Science & Machine Learning Enthusiast
Technical Skills
Programming Language:
Python, SQL, R, Java, Kotlin, Swift, C++, C#, R, html, CSS, PHP, JavaScript, MATLAB, Assembly Language
Software:
Visual Stodio Code, PyCharm, Jupyter Notebook, Notepad ++, Google Colab, MySQL, Android Studio Code, Azure Cognitive Service, AWS, Tableau, PowerBi, Weight & Biases, Microsoft Office Suite (Word, PowerPoint, Access, and Excel: VBA, pivot tables, slicers, conditional formatting), GIS
Other Skills:
Excellent people management skill, Inclusivity skill, Collaboration expertise, Conflict resolution proficiency, Detailed and organized, Capability to lead, Excellent communication skill, Ability to multitask, Project Planning, Project Engineering
Education
-
Master of Science (M.Sc.) Computer Science & Systems University of Washington Tacoma Present -
Bachelor of Science (B.Sc.) Electrical & Electronic Engineering Chittagong University of Engineering and Technology, Chattogram, Bangladesh October 2022 CGPA: 3.43/4 -
Higher Scecondary School Certificate (HSC) Science Chittagong College, Chattogram, Bangladesh March 2016 GPA: 5/5
Work Experience
Data Intern @ IB Analytica, USA (Mar 2025-May 2025)
- Assist the AI teams to develop an AI model using LLMs for investigative journalism.
- Research data sources and use APIs to fetch the data from social platforms related to the project.
- Lead a team of three interns; collaborated on improving AI usability for end users through data analysis.
Research Assistant @ University of Washington Tacoma (Jul 2024-Present)
PI: Prof. Dr. Mohammad Ali, University of Washington Tacoma
- Collaborate on research in Dynamic Neural Network Compression for Scalable AI Deployment.
- Survey on different model compression techniques such as pruning, quantization, knowledge distillation, low rank factorization, building efficient neural network to deploy the model on Edge devices.
- Gain knowledge on reducing computational cost, inference time, memory usage, and FLOPs to deploy models on mobile devices, and provided offline functionality of AI models on resource-constrained devices.
- Study different state space models, such as Mamba, for research purposes.
Voluntary Research Work @ University of Washington Tacoma (Jan 2023-Dec 2023)
PI: Dr. Abdul Salama, Machine Learning Manager, Virta Health, USA
- Used Azure Cognitive Service for text extraction.
- Used Coco datasets, Pytorch Lightning, various API and Weight & Biases Visualization to fine tune the DETR and Faster R-CNN models for object detection.
- Got practical experience with real data handling, convolutional neural network and deep learning.
- Used Weight & Biases platform to visualize the learning curves and sweeps for tuning hyperparameters.
Junior Software Engineer Intern @ Sourcetop, Inc Bangladesh (Jul 2022–Dec 2022)
- Developed web applications using Visual Studio, CSS/HTML/JavaScript, C#, Django, PHP, WordPress.
- Developed mobile applications using Kotlin, Carto, Swift, Android Studio.
- Data Analysis and Mining with Python
- SQL Querying and optimization
Research Experience
MiniMedAssistant: A Multimodal Medical Assistant Model for Resource Constrained Devices. (Master’s Thesis)
- Develop a multimodal medical assistant AI with reasoning capabilities by curating data via knowledge distillation, applying pruning and quantization for model compression, and deploying it on mobile devices with offline functionality.
Publications
Conference
Projects
Quantization Aware Dynamic Scheduling Task Parallelization for Resource Constrained Devices
- Developed a quantization-aware dynamic task scheduling framework.
- Integrated heterogeneous task parallelization to improve hardware utilization.
- Applied dynamic graph partitioning and graph coarsening to optimize model data flow.
- Reduced computational cost, latency, and energy consumption for real-time deployment.
Query Optimization: Monitoring, analysis and enforcing query plans
- Study the behavior of a query optimizer inside a database management system.
- Query plan change in response to changes in the DB schema.
- Query plan change in response to a change in the cardinalities
- The optimizer makes the wrong decisions.
Similarity Detection in Trajectory Databases to Facilitate Ridesharing
- Use QuickBundles clustering algorithm with Spatio Temporal Distance metric for trajectory similarity detection.
- Utilize San Francisco Taxi Dataset, SQL, Python, DBSCAN, Hadoop to propose a updated bus path.
Lime Interpretability
Interpreting the Swin Transformer through Lime Interpretability
Understanding how deep learning models, such as the Swin Transformer, classify objects is crucial for improving their performance and trustworthiness. This study leverages LIME (Local Interpretable Model-agnostic Explanations) to shed light on the decision-making process of Swin Transformer models. By applying LIME, we aim to identify the important features that influence the model’s classifications and gain insights into its internal workings. This interpretability analysis not only helps in debugging and refining the model but also enhances transparency, making it easier to trust and adopt these advanced models in real-world applications.
Understanding Feature Importance in Mask2Former
Feature Importance of Mask2Former
In the realm of computer vision, the Mask2Former model has emerged as a state-of-the-art framework for various tasks such as image segmentation and object detection. Understanding the feature importance in Mask2Former is crucial for interpreting its predictions, improving model robustness, and ensuring transparency in its decision-making process. This repo delves into the mechanisms of Mask2Former, highlighting the significance of different features and their impact on the model’s performance. (on-going)
SQL Projects Projects Repo: SQL Projects
Implementation of a GPS based road sign and speed breaker alarming system
Road safety is a critical concern, and effective monitoring of road signs and speed breakers is essential for preventing accidents and ensuring smooth traffic flow. This project presents the implementation of a GPS-based road sign and speed breaker alarming system designed to address the challenges associated with monitoring road signs and speed breakers. The system includes a voice command-based alarm to alert drivers in real-time, enhancing their awareness and promoting safer driving practices. Presentation
Huawei Flagship Program
Floating Network Solutions from Bangladesh Perspective
As the demand for robust and adaptable network solutions grows, floating network solutions have emerged as a vital technology to address connectivity challenges in various environments. This project explores the concept, design, implementation, and benefits of floating network solutions from Bangladesh Perspective, highlighting their potential to provide enhanced connectivity and resilience in diverse scenarios.
Training
Huawei Seeds for The Future Flagship Program, 2020
Leadership Development
General Secretary @ Andromeda Space & Robotics Research Organization, Chittagong University of Engineering & Technology, Bangladesh (Dec 2020–Mar 2022)
- Arranged different coding hackathon, webinar and robotics competition.
- Managed the committee with a focus on equity, integrity, and diversity.
- Ensured a welcoming environment for all new students.
- Maintained strong connections with faculty, alumni, and students to support the organization’s mission.
- Distributed tasks among senior and junior members to foster a sense of belonging.
- Organized webinars, seminars, and workshops on career growth and technologies.
- Collaborated with the president and financial secretary to manage the organization’s funding.
- Posted the organization’s work on the Facebook page and maintained connections with students.
- Managed and regularly updated the organization’s page as the page administrator.
- Built a Line Follow Robot with a team for a competition.
- Developed a GPS-based road sign and speed breaker alarming system utilizing Raspberry Pi technology to monitor road signs and speed breakers and implemented a voice-command-based alarm system.
Design & Decoration Secretary @ Chittagong Student Forum, CUET (Apr 2021–Mar 2022)
- Coordinated and executed office operations, ensuring efficient workflow and organization.
- Provided administrative support, including managing schedules, preparing reports, and handling correspondence.
- Assisted with the planning and execution of design and decoration projects, ensuring all tasks were completed on time and within budget.
- Posted the organization’s work on the Facebook page and maintained connections with students.
Python Instructor @ Andromeda Space & Robotics Research Organization, Chittagong University of Engineering & Technology, Bangladesh (Dec 2019–Mar 2022)
- Guided the organization team for various projects such as traffic light signal and traffic light recognition.
-
Took classes in different python libraries such as NumPy, matplotlib, OpenCV, pandas, scikit-learn, Python EDA analysis etc.
- Guided the organization team for various projects such as traffic light signal and traffic light recognition.
Hobby
- Singing and playing piano (I’ve taken part in many programs where I performed singing in various settings from small community gatherings to larger events. You might enjoy some of them!)
