Srabonti Deb
M.Sc. in Computing Science, University of Alberta (2026–Present) · B.Sc. CSE, RUET (2019–2024)
I am a Master's student in Computing Science at the University of Alberta and a Computer Science & Engineering graduate of Rajshahi University of Engineering & Technology (RUET). Previously, I worked as a Research Assistant at Young Learners' Research Lab. My research interests span deep learning, computer vision, trustworthy AI, and NLP — with a focus on real-world impact in safety-critical and low-resource domains.
Research interests — hover to explore
Experience
Work History– Jun 2026
Contributed to projects aimed at improving machine learning models across three key research tracks:
- Developed a RAG system tailored for low-resource and noisy text environments.
- Utilized BM25 (lexical keyword matching) and Contriever (dense semantic retrieval) for answer extraction.
- Integrated retrieval pipelines with generator models: Flan-T5 and Llama-3 Instruct.
- Developed diffusion-based augmentation strategies to address class imbalance in medical image datasets.
- Applied preprocessing: CLAHE, Gaussian smoothing, MSRCR, and bilateral filtering.
- Trained ConvMixer, ConvNext-Tiny, and Swin Transformer with stratified k-fold validation.
- Designed a 3D InceptionNet framework for multi-class deepfake detection on FaceForensics++ (6 categories).
- Implemented three-phase conditional training via transfer learning to improve minority class performance.
- Evaluated robustness across five manipulation techniques with comprehensive confusion matrix analysis.
– Dec 2023
- Developed frontend components and features using TypeScript, React & Tailwind CSS.
- Contributed to two client projects: an Inventory Management System (e-commerce) & Trip-Portal (travel agency website).
- Focused on responsive layouts, reusable components, and smooth user experience.
– Dec 2020
- Designed interfaces, typography, color systems, and icons using Adobe XD & Figma.
- Improved usability, accessibility, and overall interaction experience of the product.
Education
Academic BackgroundCGPA: 3.43 / 4.00
Thesis: "Advancing Deepfake Detection: Integrating Time Series Processing with Vision Transformers."
Thesis Supervisor: Tasmia Jannat
Relevant Coursework:
Linear Algebra · Differential and Integral Calculus · Data Structures and Algorithms · Object Oriented Programming · Database Management · Applied Statistics & Queuing Theory · Digital Image Processing · Neural Network and Fuzzy System · Artificial Intelligence · Data Mining
GPA: 5.00 / 5.00
Science Group · Chittagong, Bangladesh
GPA: 5.00 / 5.00
Science Group · Chittagong, Bangladesh
Publications
Research OutputProjects
Selected Work- Built an intelligent chatbot capable of answering questions across multiple uploaded PDFs using open-source NLP models.
- Utilized Sentence-Transformers (all-MiniLM-L6-v2) and LangChain with FAISS for semantic text retrieval and similarity search.
- Integrated Hugging Face LLM (Llama-3.1-8B-Instruct) with ConversationBufferMemory for context-aware, multi-turn QA.
- Developed an interactive Streamlit interface for real-time, conversational exploration of uploaded PDFs.
- Built responsive UI across devices with autocomplete search and product preview features.
- Implemented security via password hashing and securely integrated payment gateways.
- Optimized database schema with indexes for improved performance and scalability.
- Developed a user database system capturing user details and up to five trusted emergency contacts.
- Implemented a one-tap emergency feature that sends live GPS location and assistance request instantly.
Technical Skills
ToolboxAwards
RecognitionLanguages
CommunicationVolunteering
Community- Managed class schedules and facilitated communication between students and faculty across 5 academic years.
- Coordinated departmental activities, examinations, and academic events.
Others
Beyond ResearchI write about topics in machine learning, computer vision, and software development — sharing insights from research, projects, and things I find interesting along the way.