https://ar.iub.edu.bd/handle/11348/624 By CSE Department 2026-10-09T02:11:02Z https://ar.iub.edu.bd/handle/11348/1641 IUB-RAG-LLM: an extensible system for continuous knowledge ingestion and asynchronous indexing Nahiyan, Zulker Nayeen Disseminating institutional knowledge at universities is challenging, resulting in information gaps for students and increased burdens on both faculty and staff who must answer repetitive queries. We present IUB-RAG-LLM, an extensible retrieval-augmented generation system designed to provide instant access to institutional knowledge. The system features an asynchronous data pipeline that reduces knowledge update latency by 6x, and a modular architecture that enables integration of new data sources and long-term maintainability. We created a corpus of 2834 documents produced from official university channels, which is processed into 7798 semantic chunks and continuously updated. A preliminary human evaluation study with 3 student participants assessing 60 question-answer pairs demonstrated the system’s effectiveness, with average correctness and completeness scores of 3.57/5.00 and 3.67/5.00, respectively. The IUB-RAG-LLM system shows promise in alleviating faculty and staff workload while improving student access to accurate, timely institutional information. Our project can be found at https://github.com/znnahiyan/iub-rag-llm. This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science (BSc) in Computer Science and Engineering (CSC), 2025. 2025-12-01T00:00:00Z https://ar.iub.edu.bd/handle/11348/1640 Kiki: designing a socially assistive robot with Bangla augmentative and alternative communication (AAC) for children with mild to moderate (levels 1 and 2) autism Saha, Ratna; Siddiqua, Mubasshira Children with Autism Spectrum Disorder (ASD) in low-resource settings often do not have access to engaging, culturally relevant therapeutic tools. This paper introduces “Kiki,” a Socially Assistive Robot developed as a communication and educational companion for Bangla-speaking children, including level 1 and 2 ASD children (nonverbal and minimally verbal). Kiki was created in close collaboration with local therapists, with a culturally responsive approach at its core. It consists of a touchscreen Bangla Augmentative and Alternative Communication (AAC) interface inspired by the Picture Exchange Communication System (PECS). Kiki uses this interface, featuring daily quiz learning and empathetic, AI-driven interactions, to provide a supportive environment for children to build vocabulary, express emotions, and practice social communication. A thorough evaluation in several autism centers in Dhaka showed strong acceptance and engagement. Therapists praised the robot’s easy-to-use Bangla interface, while children formed a special bond with Kiki, showing increased focus and motivation during sessions. As a proof of concept made for a fraction of the cost of commercial robots, Kiki provides a caring and scalable model for helping children with ASD in diverse linguistic communities around the world. This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science (BSc) in Computer Science and Engineering (CSC), 2025. 2025-12-14T00:00:00Z https://ar.iub.edu.bd/handle/11348/1625 Breast cancer detection using hybrid convolutional neural network Akif, Ikram Hossain; Tuly, Sadia Habib This thesis evaluates frozen-backbone transfer learning for automated Normal-versus-Disease triage in Digital Breast Tomosynthesis (DBT). Using the BCS-DBT dataset, 17 architectures were tested with different slice-window depths and classifier widths, followed by hyperparameter optimization and spatial/probability fusion. Compact CNNs performed better than larger or attention-based models under the frozen constraint. While fusion improved sensitivity and F1 score, overall AUROC remained limited, demonstrating the performance ceiling of frozen transfer learning for DBT triage. This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science (BSc) in Computer Science and Engineering (CSC), 2026. 2026-08-01T00:00:00Z https://ar.iub.edu.bd/handle/11348/1619 A novel mobile-captured fabric dataset for benchmarking cnn architectures in multiclass classification Hasan, Maher This report investigates automated fabric-type identification under realistic textile production conditions using smartphone images captured without controlled lighting, backgrounds, or camera positioning. A dataset of 3,354 images covering six fabric classes—canvas, twill, denim, RFD knits, fleece, and corduroy—was developed, with areal-density measurements ranging from 186 to 371 g/m². Six ImageNet-pretrained CNN models were evaluated using a common two-stage transfer-learning approach, with and without data augmentation. EfficientNetV2-B0 achieved the best performance, reaching 99.8% accuracy and 99.9% precision, recall, and F1-score with augmentation, while MobileNetV2 provided the fastest inference at 6.2 ms per image. Grad-CAM analysis showed that the models focused primarily on fabric surfaces rather than backgrounds or measurement labels, reducing concerns about shortcut learning. Overall, the study demonstrates that lightweight, compound-scaled deep learning models can accurately classify fabric types from ordinary smartphone images under uncontrolled, real-world textile production conditions. This thesis is submitted in partial fulfillment of the requirements for the degree of Bachelor of Science (BSc) in Computer Science and Engineering (CSC), 2026. 2026-08-01T00:00:00Z