New Titles

Business Feasibility Report : Blush Marketing Agency

Hiba Javed

Submitted in fulfillment of the requirements for the degree of Master in Business Administration to the department of Management Sciences. Chapter 1: Company Overview and Services- Chapter 2: Operations and Management- Chapter 3: Industry and Market Analysis- Chapter 4: Marketing Strategy- Chapter 5: Financial Plan- Chapter 6: Risk Management- Chapter 7: Milestone and Schedule- Chapter 8: Financial Strategy- Chapter 9: Growth Strategy and Feasibility - Chapter 10: References. This is a full yearlong project report comprising of a business plan and a feasibility study of the Blush Marketing Agency (BMA) which is a small business project launched by the author of this report in Faisalabad, Pakistan under the category of a business development project. The study aims to explore the commercial viability, strategic position and growth potential of BMA in the dynamic marketing landscape of SMEs in the context of Pakistan.The study is clear, that there's a significant and consistent SME gap in the SME segment in Pakistan,where only 73% of the small and medium businesses think that their marketing strategy is correct (Constant Contact & Ascend2, 2024) leading to systemic commoditisation, margin erosion and failure to maintain competitive difference. There is a considerable proportion of SME manufacturing units and consumer product units in Faisalabad, the third largest city and industrial capital of Pakistan with good product quality, but poorly developed brand identity. It is not possible to find any boutique brand strategy agency in this market at SME friendly pricing. BMA is ready to serve the need with its comprehensive solution comprising of brand strategy, visual identity, social media management, content creation, digital advertising and brand consulting in three packages: Starter (PKR 25,000 to 35,000/month), Growth (PKR 40,000 to 60,000/month) and Premium (PKR 75,000 to 100,000/month).The financial analysis shows remarkable use of the capital: It took just Month 1 to achieve the operational break-even, the financial working capital reserve was just PKR 104,420 and the projected Year 1 net profit of PKR 2,043,000 with 57% margin. The estimated 10-year NPV (at a 15% discount rate) and IRR are PKR 20,000,000 and approximately 185%, respectively. Porter's Five Forces, SWOT analysis, PESTLE, Ansoff Matrix and detailed competitor analysis validate the reality of BMA's strategic niche in Faisalabad and its growth potential being quite promising, new service lines such as e-commerce brand management, BMA Brand Academy, and influencer marketing, and a new geographic market segment - Lahore, Karachi and Gulf diaspora.

Date Published:2026

The Role of Artificial Intelligence in Recruitment and Selection of Private Sector Organizations

Muhammad Saqlain

Submitted in fulfillment of the requirements for the degree of Master in Business Administration to the department of Management Sciences. Chapter 1:Introduction- Chapter 2: Literature Review- Chapter 3: Research Methodology- Chapter 4: Data Analysis and Results- Chapter 5: Discussions and Implications- Chapter 6: Conclusion and Recommendations. The enormous advancement of Artificial Intelligence (Al) has significantly contributed to the shift in the Human Resource Management practices, particularly recruitment and selection. Old-fashioned recruiting processes can be quite time-consuming, subjective and prone to inefficiencies and are forcing organizations to turn to AI-controlled hiring solutions to enhance efficiency. The research examines the use of Al in recruitment and selection within the privatized sector organizations specifically looking at the different notable Al functions, including Natural Language Processing capability (NLP), Machine Vision accuracy, Automation level, and Augmentation intensity. Also, the analysis is done to judge the mediating impact of organizational experience and the moderating impact of Al-readiness of the HR professionals in calculating the outcome of the recruitment and selection.The research philosophy adopted positivism research, and the research method was quantitative research based on the descriptive and correlational research design. Along with the assistance of structured questionnaire, data collection was carried out amongst the 225 human resource professionals, who were involved in the work of the private sector organizations. The data were analyzed with the help of Partial Least Squares Structural Equation Modeling (PLS-SEM) on the basis of SMART-PLS and evaluated those hypotheses. As can be seen, the AI capabilities in each instance have a significant positive effect on the recruitment and selection outcomes through organizational experience except augmentation intensity, including improving efficiency, reducing the time to hire, enhancing evaluation accuracy, and improving fairness in the selection and hiring of candidates. The most influential were NLP capability and automation level.The findings also verify that organizational experience does exist to mediate the relationship between AI capabilities and the results on the organization and the association of organization outcomes of recruitment mediate the working of AI-driven recruitment systems and enhances the effectiveness of such systems. The research bas an input into the novel literature in that there is empirical data on the synergistic effect of technological, organizational and human factors in the recruitment of Al. In practical terms, findings will inform the organizational, HR practitioners, and policymakers in their quest to undertake the implementation of Al in the hiring and selection using the hiring and selection sector responsibly and viably.With more organizations pursuing digital transformation, the findings of this research offer a practical guide for leveraging AI to enhance recruitment efficiency .

Date Published:2026

Hunar Bazar

Ayesha Nazir

Date Published:2026

Frank Wood's business accounting : An introduction to financial accounting

Sangster, Alan.

The industry-leading book in accounting methods and techniques, ideal for anyone taking a first course in financial accounting at university or college. "Essential for introductory accounting teaching for both specialist and non-specialist students. It skilfully progresses from the basics, employing insightful examples and questions for effective experiential learning." Kudz Munongi, Senior Lecturer in Finance and Accounting, University of Sunderland Frank Wood's Business Accounting, 16th edition, is known for its clear, readable style and is one of the world's best-selling textbooks on financial accounting. Ideal for those new to the field, as well as the more experienced students and professionals, this textbook will introduce you to essential ideas and methods in financial accounting - from the all-important terminology and techniques to the key financial statements. Key features include: Short, manageable chapters, allowing you to build up your knowledge step by step A wide range of in-chapter activities and end-of-chapter questions, many of which are new for this edition, enabling you to continually test your understanding Substantial updates to more than half of the 35 chapters, ensuring continued alignment with modern accounting techniques as well as incorporating an increased number of in-chapter worked examples Integrated, up-to-date coverage of relevant International Financial Reporting Standards (IFRS) Answers at the end of the bookr so you can learn as your you practice, and solidify your knowledge. Used by generations of students across the world, this hugely popular text provides a thorough and up-to-date introduction to the subject, including all the financial accounting fundamentals required by major accountancy exam bodies. "Frank Wood's Business Accounting is an easy-to-read introductory financial accounting textbook that provides numerous worked examples and end-of-chapter questions of increasing complexity, which allows the reader to build up their accounting knowledge in a logical manner." Associate Lecturer, The Open University Pair this text with MyLab®Accounting MyLab®Accounting enables students to master concepts and develop their accounting skills. By combining a range of over 775 varied questions and accounting problems with flexible tools to help students at point of need and short videos on crucial concepts, MyLab engages students and improves results. MyLab® Accounting is not included with this title. If you would like to purchase both the physical textbook and MyLab®Accounting (which also comes with the eBook), search for: 9781292459578 Frank Wood's Business Accounting, 16th edition 'MyLab via Bundle', which consists of: Print textbook eTextbook MyLab®Accounting Students, MyLab® should only be purchased when required by an instructor. If MyLab is a recommended/mandatory component of the course, please check with your instructor for the correct ISBN. Instructors, contact your Pearson representative for more information.

Date Published:2025

Couturio

Huda Noor

Date Published:2026

DRIP ----- 360 Degree : Try Before You Buy.

Aneeqa,Khan

Submitted in partial fulfilment of the requirement for the degree of Bachelor of Science (Electrical Engineering) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Background 1.2 Motivation 1.2.1 Accessibility in Research 1.2.2 Environmental Responsibility 1.3 Project Goals 1.4 Project Scope 2 Problem Statement and United Nations Sustainable Development Goals 2.1 Problem Statement 2.2 United Nations Sustainable Development Goals 2.2.1 Rank of Pakistan in SDGs 3 Literature Review 3.1 The Evolution of Virtual Try-On Systems 3.1.1 An Overview of the Early GAN-Based VTON Systems (2017 to 2020) 3.1.2 Innovative Progression in the Diffusion-Based Techniques (2022-2024) 3.2 Multi-view Consistency and 3D Representations 3.2.1 NeRF-Based Editing 3.2.2 Three-Dimensional Gaussian Splatting 3.2.3 VTON 360: The Underlying Infrastructure 3.3 Memory-Efficient Attention Mechanisms 3.3.1 XFormers Library 3.3.2 Flash Attention 3.4 Evaluation Criteria for Image Generation 3.4.1 SSIM Metric 3.4.2 LPIPS 3.4.3 CLIP and DINO Similarity 3.5 Research Gap 3.5.1 Research Gaps Identified 3.5.2 Our Contributions to the Research Gaps 4 Methodology 4.1 System Overview 4.2 Step 1: Inference from Multi-View Try-On 4.2.1 Preprocessing of Input 4.2.2 Diffusion Pipeline Architecture 4.2.3 Core Innovation: Segmented Multi-View Attention 4.2.4 Inference Configuration 4.3 Stage 2: Human Parsing Accelerated by ONNX 4.3.1 Model Architecture 4.3.2 ONNX Optimisation Pipeline 4.3.3 Inference Pipeline 4.4 Stage 3: Refinement Through Identity Preservation 4.4.1 Mask Creation 4.4.2 Stability Widening for the Non-Identifying Image Framework 4.4.3 Gaussian Feathering 4.5 Stage 4: Package 3DGS 4.5.1 Data Structure and Configuration 4.5.2 Packaging Script 4.5.3 GPU-Based External 3DGS Training 4.6 Evaluation Framework 5 System Implementation 5.1 Technology Stack 5.2 Directory Structure 5.3 Operating the Pipeline 5.4 Management of Configuration 6 Results and Analysis 6.1 Dataset 6.2 Pipeline Stages and Visual Workflow 6.2.1 Garment Input Specifications 6.2.2 Multi-View Input and Geometric Guidance 6.2.3 Person-Agnostic Representation 6.2.4 Semantic Segmentation and Parsing 6.2.5 Final Refined Results 6.3 Qualitative Results on Alternate Identities 6.4 Evaluation Protocol 6.5 Quantitative Results 6.6 Qualitative Analysis 6.7 Ablation Studies 6.8 Comparison with Baselines 7 Discussion of Results 7.1 Memory-Efficiency Tradeoffs 7.1.1 94% Memory Savings with Reduced Multi-View Consistency 7.2 CPU and GPU: Which One Should You Choose? 7.2.1 Advantages of CPU Deployment 7.2.2 Advantages of GPU Deployment 7.2.3 Use Case Matrix 7.3 Limitations 7.3.1 Inference Latency 7.3.2 Sparse View Consistency 7.3.3 Garment Type Limitations 7.3.4 Batch Processing Constraints 7.3.5 Manual Checkpoint Initialization 7.4 Extending to Other Datasets 7.4.1 MVHumanNet Compatibility 7.4.2 Custom People Data 8 Conclusions 8.1 Summary of Contributions 8.1.1 Technical Contributions 8.1.2 Evaluation Framework 8.1.3 Open Source Contributions 8.1.4 Project Objectives Accomplishment 8.2 Effects and Implications 8.2.1 Academic Implications 8.2.2 Industry Implications 8.2.3 Environmental Implications 8.3 Lessons Learned 8.3.1 Technical Observations 8.3.2 Process Observations 9 Recommendations for Future Work 9.1 Short-Term Improvements (3-6 Months) 9.1.1 Hybrid CPU-GPU Scheduling 9.1.2 INT8 Quantization 9.1.3 View Count Optimization 9.1.4 Towards Real-Time Inference 9.2 Medium-Term Research Areas (6-12 Months) 9.2.1 Diffusion Model Knowledge Distillation 9.2.2 Physical Garment Dynamics 9.2.3 Mobile Deployment via Neural Accelerators 9.2.4 Multi-Garment Composition 9.3 Future Outlook (1-2 Years) 9.3.1 Unified CPU-GPU Architecture 9.3.2 IVTO Foundation Model 9.3.3 Augmented Reality Social Platforms Bibliography A Detailed System Specifications A.1 Development Machine Infrastructure A.2 Software Dependencies and Libraries B Details of Formulation for Metrics B.1 Derivation of Structural Similarity Index (SSIM) B.2 Formulation of Learned Perceptual Image Patch Similarity (LPIPS) C Configuration Files C.1 Complete Multi-View Inference Configuration D Sample Output E Source Code Listings E.1 Stage 1: Multi-View Try-On Inference E.1.1 Dataset Initialization E.1.2 Image Transformation Pipeline E.1.3 Base Model Loading E.1.4 Segmented Multi-View Attention Processing E.1.5 View Weight Matrix Construction E.2 Stage 2: ONNX-Accelerated Human Parsing E.2.1 ONNX Session Configuration E.2.2 ONNX Forward Pass and Output Merging E.2.3 Post-Processing Automation Script E.3 Stage 3: Identity-Preserving Refinement E.3.1 Binary Mask Generation E.3.2 Morphological Dilation Processing E.3.3 Non-Identifying Image Stability Check E.3.4 Gaussian Feathering Mechanism E.3.5 32-Bit Floating-Point RGB Composite Blending E.4 Stage 4: 3DGS Packaging E.4.1 Directory Structure and Camera Transforms Schema E.4.2 Package Structural Validation Pipeline E.4.3 Automated Structural Asset Deployment E.4.4 3DGS Training and Novel View Rendering Pipeline E.5 Evaluation Framework E.5.1 Evaluation Execution Script and Structured Output Schema. While the VTON360 framework provides an excellent virtual try-on experience, it currently requires a large amount of GPU VRAM. We have created a new CPU-optimized pipeline (based on a highly memory-efficient method for attention). This work is intended as a VI prototype for experimental use, but plans to develop it into a production-level system are being considered. As a result of this work, we reduced peak RAM usage by 94% (from 25 GB to 1.5 GB) while maintaining 75%-85% visual output quality (SSIM of 0.85) and producing high-quality multi-view virtual try-on results. It generates 4 cardinal views for 3D Gaussian splatting in NeRFStudio, which demonstrates that this method is feasible and provides high-quality virtual try-on experiences on consumer hardware.

Date Published:2026

Marketing management

Kotler, Philip.

textbooks. La nouvelle édition du manuel de référence sur le marketing ! Cet ouvrage de référence, qui en est à sa 17e édition, présente tous les concepts et outils du marketing de manière vivante et pédagogique, tout en intégrant les avancées les plus récentes. Il est soutenu par de très nombreux exemples, des cas d'entreprises, des approfondissements conceptuels et méthodologiques. Les photos présentent des campagnes publicitaires, des points de vente et des produits témoignant d'un marketing dynamique et innovant. Cette édition expose comment construire et mettre en oeuvre une politique marketing en phase avec les réalités actuelles des marchés et des entreprises. Elle prend en compte les enjeux de consommation responsable, les conséquences environnementales et sociétales du marketing, les progrès de la technologie, l'intelligence artificielle, les données de masse, le rôle central des réseaux sociaux. Cette nouvelle édition se caractérise également par : - une structure remaniée avec de nouveaux cas sur des entreprises et marques innovantes comme Action, ChatGPT, Yuka, Back Market, HiPRO, Skims, Basic-Fit, Mistral AI, ou sur des références emblématiques comme Louis Vuitton, Canal+, Nespresso, SNCF, Relais & Châteaux, Lego, Spotify... ; - de nombreux encadrés détaillant l'impact de l'IA sur les pratiques du marketing ; - de nouveaux encadrés pour approfondir certains sujets : les nouvelles tendances de l'emballage, la publicité programmatique, l'intrusion publicitaire, ou encore la gestion des émotions en marketing B-to-B ; - des développements fondés sur les dernières avancées de la recherche : le ciblage comportemental en ligne, le démarketing, l'innovation durable, le rôle du marketing dans les transitions environnementales et sociétales...»--Quatrième de couverture.

Date Published:2026

Mira : A multi step internet reasoning agent for accessible web automation

Maham Ismail

Date Published:2026

Early Diagnosis of Autism Using YOLOV11 : A Comparative Study with YOLOV8

Saman,Siddiqui

Submitted in partial fulfilment of the requirement for the degree of Bachelor of Science (Electrical Engineering) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Background of the Study 1.2 Problem Background 1.3 Research Objectives 1.4 Research Questions 1.5 Project Overview 1.6 Scope and Delimitations 1.7 Thesis Organization 2 Problem Statement and SDGs 2.1 Problem Statement 2.2 Research Contributions 2.3 Mapping to Sustainable Development Goals (SDGs) 2.4 Significance of the Study 3 Literature Review 3.1 Overview of Autism Diagnosis 3.2 Machine Learning in ASD Detection 3.3 Deep Learning and Computer Vision 3.3.1 YOLO Architectures 3.4 Research Gaps 4 Methodology 4.1 Introduction 4.2 Dataset Development 4.2.1 Data Sources 4.2.2 Dataset Composition and Class Distribution 4.3 Data Preparation Pipeline 4.3.1 Data Annotation 4.3.2 Preprocessing 4.3.3 Data Augmentation 4.3.4 Dataset Splitting 4.4 System Architecture 4.4.1 Proposed Framework 4.4.2 Baseline Architecture (YOLOv8) 4.4.3 Enhanced Architecture (YOLOv11) 4.5 Experimental Setup 4.5.1 Hardware and Software 4.5.2 Training Hyperparameters 4.6 Evaluation Metrics 5 Results and Analysis 5.1 Introduction 5.2 Training Dynamics 5.2.1 Loss Convergence 5.2.2 Metric Progression 5.3 Quantitative Evaluation 5.3.1 Overall Model Performance 5.3.2 Confusion Matrix Analysis 6 Discussion of Results 6.1 Interpretation of Quantitative Results 6.1.1 Architectural Analysis 6.1.2 Comparison with the Base Paper 6.2 Limitations 7 Conclusion 7.1 Summary of the Study 7.2 Synthesis of the Research Findings 7.3 Fulfillment of the Research Objectives 7.4 Significance and Social Impact 7.5 Concluding Remarks 8 Recommendations for Future Work 8.1 Future Work 8.2 Technical Recommendations 8.2.1 Real-Time Edge Detection 8.2.2 Temporal and Video-Based Detection 8.3 Clinical and Multimodal Suggestions 8.3.1 Multimodal Data Integration 8.3.2 Longitudinal Studies Bibliography A Source Code A.1 Dataset Configuration (data.yaml) A.2 Training Script A.2.1 YOLOv8 A.2.2 YOLOv11 A.3 Validation A.4 Test the Model. Autism Spectrum Disorder (ASD) is a neuro-developmental condition affecting com-munication, behavior, and social interactions. Traditional diagnostic methods rely on prolonged clinician observation, making them subjective and difficult to scale and can be burdensome for parents in Pakistan. This creates a need for more accessible objective screening tool. This research evaluates the efficacy of deep learning-based facial expression recognition as a supportive tool for early ASD detection. This research presents a comparative analysis between YOLOVS, serving as the baseline and the latest YOLOv11 architecture. Both models were trained to recognize seven facial expressions using a dataset curated, annotated, and, augmented via Roboflow. The results demonstrate that while YOLOV8 establishes a reliable foundation for facial pattern recognition YOLOv11 provides significant improvements in accuracy. By benchmarking these two architectures this study highlights the real potential of evolving real-time object detection models to reduce diagnostic delays and support inclusive education and healthcare systems.

Date Published:2026

Sam : The twin

Shaheera Imtiaz

Submitted in fulfillment of the requirements for the degree of Bachelor Science in Computer Science to the department of Computer Science. Chapter 1: Vision Document- Chapter 2: Stakeholder and User Description- Chapter 3: System Features- Chapter 4: Functional Requirements- Chapter 5: Non Functional Requirements- Chapter 6: Use Case Models- Chapter 7: System Diagram- Chapter 8: Testing- Chapter 9:References. SAM-The Twin is a virtual assistant for students and introverts. In contrast to other applications where you could use any function separately, for example, making notes or keeping a journal, here everything is integrated into one. SAM will assist you in your tasks, mood tracking, memories storing, and many more. With features like a safe memory vault, venting mode, tools for decision-making, adaptive communication, and positive reinforcement, the software fosters mental health in addition to increasing productivity. SAM-The Twin, which was created with the aid of contemporary tools like Firebase, Android Studio, Figma, Visual Studio Code, and Firebase ML Kit, guarantees both security and functionality. The application's ultimate goal is to give users a reliable digital twin that gives them the ability to feel heard, supported, and in charge of their lives.

Date Published:2026

AI-Powered Robotic Arm for Automated Defect Detection and Sorting of Bottled Products

Azeem,Asif

Submitted in partial fulfilment of the requirement for the degree of Bachelor of Science (Electrical Engineering) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Background 1.1.1 Process Flow Architecture 1.2 Motivation 1.3 Objectives 1.4 Scope and Limitations 1.5 Project Structural Specifications 2 Problem Statement and Sustainable Development Goal 2.0.1 SDG 9: Industry, Innovation, and Infrastructure 3 Literature Review 3.1 Electromechanical and PLC Foundations 3.2 Evolution of Computer Vision: From OpenCV to Deep Learning 3.3 The YOLO Paradigm and YOLOv8 3.4 Robotics and 6-DOF Kinematics 3.5 Gap Analysis: The Unaddressed SME Need 4 Methodology 4.1 Identifying Constraints in the Pakistani SME Manufacturing Sector 4.2 Software Architecture: Separation of Concerns 4.2.1 System Block Diagram 4.3 Hardware Integration: Karachi Component Environment 4.4 PWM Signal Integrity and Kinematic Resolution 4.5 Vision Intelligence: YOLOv8 Inference Pipeline 4.6 End-to-End Latency Audit 4.7 Mechanical Reliability Evaluation 4.7.1 Environmental Stress Characterisation 4.7.2 Economic Viability Analysis 5 Results and Analysis 5.1 Experimental Evaluation Framework 5.1.1 Training Data Distribution and Convergence Metrics 5.1.2 Classification Performance 5.1.3 Validation Batch Performance Visualization 5.1.4 Empirical System Detection Robustness Curves 5.1.5 Physical Mechatronic Assembly and Real-Time Dashboard Integration 6 Discussion of Results 6.1 Operational Observations and Systemic Trends in Live Production Runs 6.2 Statistical Precision, Recall, and the Challenge of Geometric Deformation 6.3 Kinematic Chain Optimization and Mechatronic Repeatability Profile 6.4 Reality Check: Project Objectives vs. Industrial Measurement 6.4.1 The Specular Highlight-Shadow Problem: Environmentally Engineered Systems 7 Conclusions 7.0.1 Barriers to Technology Adoption 7.1 Detection Model Summary 7.1.1 Robotic Arm Design 7.1.2 Master-Slave Control Scheme 8 Recommendations for Future Work 8.1 Industry 4.0 Transition Pathway 8.1.1 Torque Requirement Analysis 8.2 Advanced Computer Vision: Hyperspectral Imaging 8.3 Conveyor Synchronisation 8.4 Active Learning Loop 8.5 IoT Cloud Telemetry 8.6 Industrial Safety Compliance 9 Bibliography A Raspberry Pi Controller: Python Source Code B Arduino Servo Controller: C++ Source Code. Macfacturing and disfumer goods are dependent, and impulsin to achieve aleato stay in quality control was then put od t bo manually vimally icind try and try the Inmune, which intensive fatig amd personal error. The problem terpecially awwee in the small and (SME) beverage and bottling industry of developing countries like Pakistan, whose the cost of old and imported automation waitne continue relying unalfoedade Forcingerprin marmal pr In order to address the technological and economic gap, this project introduces the design, development, and experimental testing of an end-to-end mechatro solution: An Al-powered Robotie Am for Automated Defect Detection and Sorting of Bottled Products. The systern in an explicitly enalded system with conto Valtel Natious Sustaluable Development Goal 9 (Industry, Innovations and Infrastructure) that is correlated to computer vision, edge computing and soults-jotat sarchanical actuation in a single low cost platform. which is made up of YOLOvs deep learning framework and edge-computing Raspberry The system architecture is based on a high throughput visual inference systres, Pi 4 (8 GB) platform. The vision model is designed to classify the integrity of product in six different defect classes: Full, Underfillesd, Overfilled, Cap Press, Label Present, and Deformed, all in one. A rule-based logic engine is used to evaluate the outpute of the models per frame in less than 10 milliseconds after inference is cotupletel, providing a binary verdict of pass or reject. To deal with non-conforming unita without stopping the production line, a toulti-threaded software design aynchronizes the vision pipeline with a 6-DOP robotic arm controlled by an Arduino Uno and an Adafruit PCA9685 PWM driver. Interpolated, current-controlled motion control system reduces voltage fluctuations and eliminates mechanical backlash. A localized, web dashboard, accessible via the network and equipped with real-time video streaming, manual servo jogging and named-pose recording, belts with operational monitoring and spatial calibration. In conclusion, this project provides a highly reproducible and open-source paradigm for automation completed at a total bill of-materials price of less than PKR 35,000, which is a viable commercial automation solution, and provides a high return alternative to traditional industrial vision systems.

Date Published:2026

4-in-1 IMU-Based Cursor and Gesture Control Device

Khan,Jaleed Abdur Rehman

Submitted in partial fulfilment of the requirement for the degree of Bachelor of Science (Electrical Engineering) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Overview 1.2 Motivation 1.3 Project Objectives 1.4 Scope and Contributions 1.5 Thesis Organization 2 Problem Statement and United Nations Sustainable Development Goals 2.1 Problem Statement 2.2 Alignment with the United Nations Sustainable Development Goals (SDGs) 2.2.1 SDG 9: Industry, Innovation, and Infrastructure 3 Literature Review 3.1 Introduction 3.2 The Air Mouse Paradigm and Inertial Limitations 3.3 High-Precision Stylus Tracking and the Importance of Optical Flow 3.4 Embedded Gesture Recognition 3.5 3D Navigation and the Space Mouse Paradigm 3.6 Summary and Research Niche 4 Methodology 4.1 System Architecture Details 4.1.1 Introduction to the Distributed Microcontroller Paradigm 4.1.2 The Mode Matrix 4.2 Hardware Selection and Integration Logic 4.2.1 Component Selection 4.2.2 Hardware Interfacing 4.2.3 Power Management Subsystem 4.3 Engineering Constraints and the Necessity of Partitioning 4.3.1 Internal DRAM Exhaustion and Heap Fragmentation 4.3.2 Radio Transceiver (PHY) Timeslicing Contention 4.3.3 Software Complexity and Core Starvation 4.4 Communication Layer and Data Pipelines 4.4.1 Communication Protocol Rationale 4.4.2 End-to-End Data Pipelines 4.5 Hardware Design and Implementation 4.5.1 Circuit Topology and Component Mapping 4.5.2 Carrier Board Architecture 4.5.3 Bus Optimization and Signal Integrity 4.5.4 Mechanical Integration and Ergonomics 4.5.5 PCB Layer Stack and Component Distribution 4.5.6 PCB Layout and Physical Constraints 4.5.7 Mechanical Layout and Mounting 4.6 Enclosure Design 4.6.1 Digital Twin and Component CAD Modeling 4.6.2 Component-Level Breakdown and Assembly 4.6.3 Manufacturing Considerations 4.7 Firmware and Application Methodology 4.7.1 Air Mouse Implementation (Gyroscopic Navigation) 4.7.2 Space Mouse Implementation (3D Manipulation) 4.7.3 Optical Surface Tracking (Pen Mode) 4.7.4 Artificial Intelligence Gesture Recognition Pipeline 4.8 System Integration and Resource Management 4.8.1 Dual-ESP32 Hardware Partitioning 4.8.2 State Machine and Mode Switching Algorithm 4.8.3 Hardware Interaction: Air Mouse Mode 4.8.4 Hardware Interaction: Pen Mode 4.8.5 Hardware Interaction: Space Mouse Mode 4.8.6 Hardware Interaction: AI Gesture Mode 5 Results and Analysis 5.1 Introduction to Testing Methodology 5.2 Air Mouse and Space Mouse Performance 5.3 AI Gesture Recognition Accuracy 5.4 Pen Mouse Modality Tracking Accuracy 6 Discussion of Results 6.1 Analysis of IMU Performance and Mitigation Strategies 6.2 Viability of TinyML on Resource-Constrained Hardware 6.3 Optical Tracking Limitations and DPI Scaling 7 Conclusions 8 Recommendations for Future Work Bibliography A Bill of Materials (BOM) and Financial Analysis A.1 Per-Unit Production Cost A.2 Total R&D and FYP Project Cost B Testing Suite Source Code (Tkinter) Index. Specialized peripheral devices like remote presenters, digital graphics tablets and 3D CAD controllers are essential for modern professional and creative workflows, but they create clutter in setup, decreased portability, and higher monetary prices. Most of the previous attempts to combine all these interaction modalities in one portable device have been based on Inertial Measurement Units (IMUs). But, when using the double integration of accelerometer data (dead reckoning), there is always an integration drift that grows unbounded. This makes IMUs unusable for fine tasks like digital handwriting. This study presents a multi-modal 4-in-1 device for over-coming this significant limitation, which adapts seamlessly between an air pointer, an optical stylus, a spatial gesture controller and a 6-Degree-of-Freedom (6-DOF) space mouse, and, therefore, is cost effective. To accomplish this, a novel Distributed Microcontroller Paradigm was designed based on a dual-ESP32 architecture that ef-fectively separates the critical sensor fusion tasks from the wireless networking use of scarce memory resources, thus eliminating memory fragmentation and radio times licing problems. The hardware associates an MPU6050 IMU to track its orientation Specialized peripheral devices like remote presenters, digital graphics tablets and 3D CAD controllers are essential for modern professional and creative workflows, but they create clutter in setup, decreased portability, and higher monetary prices. Most of the previous attempts to combine all these interaction modalities in one portable device have been based on Inertial Measurement Units (IMUs). But, when using the double integration of accelerometer data (dead reckoning), there is always an integration drift that grows unbounded. This makes IMUs unusable for fine tasks like digital handwriting. This study presents a multi-modal 4-in-1 device for over-coming this significant limitation, which adapts seamlessly between an air pointer, an optical stylus, a spatial gesture controller and a 6-Degree-of-Freedom (6-DOF) space mouse, and, therefore, is cost effective. To accomplish this, a novel Distributed Microcontroller Paradigm was designed based on a dual-ESP32 architecture that ef-fectively separates the critical sensor fusion tasks from the wireless networking use of scarce memory resources, thus eliminating memory fragmentation and radio times licing problems. The hardware associates an MPU6050 IMU to track its orientation in an unconstrained environment with a PMW3610 high precision optical sensor to perform drift-free relative surface tracking. The experimental results show strong performance, with the custom Edge Impulse neural network achieving an accuracy.

Date Published:2026

Vision AI Driven Accident Risk Prediction Using Vehicle Behaviour Analysis

Hashmi,Shayan

Submitted in partial fulfilment of the requirement for the degree of Bachelor of Science (Electrical Engineering) at the National University of Computer and Emerging Sciences. 1 Introduction 1.1 Motivation 1.2 Problem Statement 1.3 Objectives 1.4 Key Contributions 1.5 Report Organisation 2 Problem Statement and United Nations Sustainable Development Goals 2.1 Problem Statement 2.1.1 Target Beneficiaries 2.2 United Nations Sustainable Development Goals 2.2.1 SDG 3: Good Health and Well-Being (Target 3.6) 2.2.2 SDG 9: Industry, Innovation and Infrastructure (Targets 9.1 and 9.5) 2.2.3 SDG 11: Sustainable Cities and Communities (Target 11.2) 2.2.4 SDG 17: Partnerships for the Goals 2.2.5 Rank of Pakistan in SDGs 3 Literature Review 3.1 Vision-Based Accident Detection 3.2 The YOLO Family of Object Detectors 3.3 Multi-Object Tracking 3.4 Vehicle Behaviour Analysis and Risk Prediction 3.5 Dataset Collection and Annotation Methodologies 3.6 Comparative Summary and Research Gaps 4 Methodology 4.1 Dataset Collection: Karachi Traffic Police CCTV Collaboration 4.1.1 Collaboration Process and Data Acquisition 4.1.2 Dataset Statistics 4.1.3 Data Privacy and Ethical Considerations 4.2 Dataset Annotation and Preprocessing on Roboflow 4.2.1 Roboflow Platform Overview 4.2.2 Frame Extraction and Selection 4.2.3 Annotation Process and Label Classes 4.2.4 Dataset Augmentation 4.2.5 Dataset Split 4.3 YOLOv11: Architecture and Training 4.3.1 YOLOv11 Architecture Overview 4.3.2 Transfer Learning from Pretrained Weights 4.3.3 Training Configuration 4.4 ByteTrack Multi-Object Tracking 4.4.1 Why Tracking Is Necessary 4.4.2 ByteTrack Algorithm 4.4.3 Trajectory Buffer 4.5 Behavioural Feature Extraction 4.5.1 Perspective Calibration and Metric Coordinate System 4.5.2 Speed and Acceleration Estimation 4.5.3 Lane Detection 4.5.4 Lane Deviation Detection 4.5.5 Time-to-Collision (TTC) Computation 4.6 Risk Scoring Engine 4.6.1 Per-Vehicle Risk Score Computation 4.6.2 Risk Zone Classification 4.7 Output Visualisation 4.7.1 Persistent Vehicle Identification 5 Results and Analysis 5.1 Hardware and Software Environment 5.2 YOLOv11 Detection Performance 5.3 Training Curves and Convergence 5.4 Component-Level Analysis 5.4.1 ByteTrack Tracking Performance 5.4.2 Lane Detection Performance 5.4.3 TTC Computation Accuracy 5.5 End-to-End Real-Time Performance 5.6 Qualitative Evaluation on Karachi CCTV Footage 5.7 Failure Mode Analysis 6 Discussion of Results 6.1 Detection Performance in Context 6.2 Pipeline Integration and Behavioural Analysis 6.3 Real-Time Performance vs. Expectations 6.4 Alignment with Research Objectives 6.5 Comparison with Related Work 7 Conclusions 7.1 Summary 7.2 Key Findings 7.3 Limitations 8 Recommendations for Future Work 8.1 Overview 8.2 Learning-Based Risk Prediction 8.3 Deep Learning Lane Detection 8.4 Nighttime Image Enhancement 8.5 Dataset Expansion and Public Release 8.6 Edge Deployment 8.7 Multi-Camera Fusion 8.8 Integration with Traffic Management Systems Bibliography A Mathematical Equations A.1 Perspective Homography Transformation A.2 Vehicle Speed Estimation A.3 EMA Speed Smoothing A.4 Longitudinal Acceleration A.5 Lateral Velocity A.6 Time-to-Collision A.7 Per-Vehicle Risk Score B Training Configuration and Risk Scoring Parameters B.1 YOLOv11 Training Configuration B.2 Risk Scoring Engine Parameters B.3 Abbreviations and Nomenclature. Road traffic accidents leading cause of fatalities worldwide, and particularly severe in rapidly urbanising cities of South Asia such as Karachi, Pakistan. The het erogeneous and often undisciplined nature of Pakistani traffic-mising private cars, motorcycles, heavy trucks, and auto-rickshaws on the same road with minimal lane discipline creates a uniquely challenging environment for conventional accident de tection systems. This research presenta Vision-Al driven framework for proactive, real-time accident risk prediction through comprehensive vehicle behaviour analysin. A novel dataset was assembled through formal collaboration with the Karachi Traffic Police, obtaining real CCTV surveillance footage captured from operational road cameras across multiple urban locations. This footage carefully curated, annotated, and labelled using the Roboflow platform, producing a high-quality la-belled dataset of four vehicle classes relevant to the South Asian context: Car, Motorcycle, Truck/Bus, and Rickshaw. A state-of-the-art YOLOvilm (You Only Look Once version 11, medium variant) object detection model was then fine-tuned on this custom dataset using pretrained weights as a starting point, achieving strong detection performance across all classes. The detection output is fed into a ByteTrack multi-object tracker that assigns and maintains persistent vehicle identities across video frames. A comprehensive behavioural analysis module then extracts safety-critical metrics from vehicle tra-jectories, including real-world speed and acceleration (via perspective homography calibration), lane occupancy and deviation (via Hough Line Transform-based lane detection), and Time-to-Collision (TTC) between vehicles sharing the same lane. These multi-dimensional behavioural signals are synthesised into a continuous risk score on a scale of 1 to 10, displayed in real time as an annotated video overlay. Risk levels of 7 and above trigger a Critical alert, designed to provide a 3-5 second lead time for intervention. The results demonstrate that combining a domain-specifically trained YOLOv11m detector with a real-time behavioural analysis pipeline produces a practical, deploy-able system for proactive traffic safety monitoring. This work contributes both a locally-sourced Pakistani traffic dataset and a complete accident risk prediction pipeline applicable to next-generation Advanced Driver-Assistance Systems (ADAS) and intelligent traffic management infrastructure. Keywords: Accident Risk Prediction, Vision-AI, YOLOvIim, Byte Track, Vehicle Behaviour Analysis, Time-to-Collision, Lane Detection, Roboflow, Pakistani Traffic Dataset, Intelligent Transportation Systems, ADAS.

Date Published:2026

Virtual Heritage

Muhammad Husnain Sabir

Date Published:2026

Strategic management : concepts and cases : a competitive advantage approach

David, Fred R.,

"This college textbook shows how to gain and sustain a competitive advantage in today's complex business world. The text helps you develop your own cutting-edge strategy through skill-developing exercises. It also offers coverage on issues related to business ethics, social responsibility, global operations, and sustainability"-- The 18th Edition features updated research, cases and examples. As a result, you'll be able to effectively create and implement a plan that can lead to a sustainable competitive advantage for any type of business.

Date Published:2025

Nexa Learn : AI powered adaptive learning and assessment platform

Eman Nabi

Submitted in fulfillment of the requirements for the degree of Bachelor Science in Computer Science to the department of Computer Science. Chapter 1: Vision Document- Chapter 2: Tools and Technologies- Chapter 3: Software Requirement Specification- Chapter 4: System Architecture and Specification- Chapter 5: High Level Diagram- Chapter 6: Comparison Table- Chapter 7: Project Time line- Chapter 8: Vision Document- Chapter 9: System Requirement Specification- Chapter 10: Use Case Diagram- Chapter 11: High Level Use Cases- Chapter 12: Expended Use Cases- Chapter 13: Testing- Chapter 14: References. "NexaLearn" is an adaptive learning and assessment tool powered by Al, which aims to change the paradigm in computer science instruction by personalizing the whole process of teaching the individual learner. Unlike many of its competitors, NexaLearn offers personalized learning opportunities via resources such as notes, videos, and exercises that are personalized based on the learner's performance. An algorithm monitors the learner's progress and helps identify gaps in learning as well as identify the learners who are at risk. In addition to that, NexaLearn meets the needs of various types of learners, including readers, video viewers, and practitioners by gradually proceeding to more advanced material.

Date Published:2026

Research Assistant

Barira Auranzaib

Date Published:2026

Smart Ads

Marwa Najeeb

Submitted in fulfillment of the requirements for the degree of Bachelor Science in Computer Science to the department of Computer Science. Chapter 1: Vision Document- Chapter 1.2: Problem Statement- Chapter 1.2: Business Opportunities- Chapter 1.3: Objectives- Chapter 1.4: Scope- Chapter 1.5: Constraints- Chapter 1.6: Stakeholder and user Description- Chapter 2: High Level Diagram- Chapter 3: System Requirement Specification- Chapter 4: Use Case Diagram- Chapter 5: High Level Use Cases- Chapter 6: Expanded Use Cases- Chapter 7: Activity Diagram- Chapter 8: Data Flow Diagram- Chapter 9: Package Diagram- Chapter 10: Unit Testing- Chapter 15: References. Adverts (ads) have become an essential part of modern marketing, assisting companies in product promotion, brand building, and social media engagement. Nowadays, they perform an important role in enhancing visibility and reaching the audience in competitive digital space. Nevertheless, it is often problematic for small businesses to create professional-looking ads due to insufficient funds, lack of knowledge in design, and the need to use multiple expensive software. Consequently, it becomes complicated for small businesses and Small and Medium• sized Enterprises (SMEs) to remain prominent in digital space and compete with big companies. In order to overcome these obstacles, SmartAds aims to develop an AI-assisted web platform, combining all the required advertising tools in one web-based solution. The platform will be capable of automated creation of logo, poster, video ad, captions, and voiceover, including custom templates and social media scheduling features. Thanks to the integration of advanced technologies such as FastAPI, OpenAI, and MoviePy, the system will provide a powerful, cost• effective, and user-friendly solution for advertisers.

Date Published:2026

Outfit Orbit

Sharaz Masih

Date Published:2026

Signity

Hamna Asif

Submitted in fulfillment of the requirements for the degree of Bachelor Science in Computer Science to the department of Computer Science. Chapter 1: Introduction- Chapter 2: Vision Document- Chapter 2.1: Problem Statement- Chapter 2.2: Business Opportunities- Chapter 2.3: Objectives- Chapter 2.4: Scope- Chapter 2.5: Constraints- Chapter 2.6: Stakeholder and user Description- Chapter 3: Software Requirements Specification- Chapter 3.1: System Feature- Chapter 3.2: Functional Requirements- Chapter 3.3: Non Functional Requirements- Chapter 4: Design Phase- Chapter 5: Test Cases- Chapter 6: References. The accessibility of video is a major problem in Pakistan, and directly affects an estimated 10 million Deaf and Hard-of-Hearing (DHH) people, who often experience educational disparities and social isolation when faced with communication barriers. Culturally irrelevant, written captioning or digital avatars are often too expensive and lack natural human expressions. Signity is an innovative, cost-effective, web-based platforms designed to solve this problem by converting spoken English and Urdu audio to Pakistani Sign Language (PSL) and developed under computer science department of F AST-NUCES Chiniot-Faisalabad Campus. The platform, featuring sequential playback of pre-recorded videos of human actors, provides a culturally representative, highly expressive and inclusive viewing experience for the educational sector, content creators and NGOs. Signity's architecture is designed to combine the power of a powerful processing engine with the high power of external language technologies. If the user uploads an MP4 video or gives a link to a YouTube video, the audio track is extracted and standardized using the FFmpeg utility. Speech-to-text transcription is performed using OpenAl's Whisper pipeline, which is specially engineered for business transcription of multilingual (English-Urdu) content and to ensure accurate word-level timestamps. The transcribed text is then sent through a special three-layer Natural Language Processing (NLP) pipeline to expand contractions, remove filler words, perform part-of-speech tagging and perform rule-based grammar transformations to restructure the text into standard PSL gloss format. At last, the system converts these gloss tokens into an internal PSL Sign Dictionary to seamlessly play the human video clips back in sync with their original time line, with a letter-by-letter fingerspelling fallback mechanism for words it doesn't recognize.Signity has broken its ecosystem down into easily navigated dashboards for public consumers and for internal signity-ers for the highest level of data integrity and community alignrnent.. The web platform has been developed with the latest version of React.js, Node.js and a safe PostgreSQL database, and it offers end-users a responsive dashboard, where they can see their processing jobs, their translation history, and can set up custom play, pause and speed options for full control of dual screen playback. At the same time, the powerful Admin Management Panel gives administration the ability to track in actual time system analytics, audit translation histories, control the fundamental sign dictionary, and review the automated low-confidence video snippets and more user-flagged video snippets. With robust security measures, such as role-based access control and input sanitization, along with rate limiting, Signity is a fitting example of a scalable enterprise that evokes a strong sense of social impact, and is directly contributing to dig ital inclusion in Pakistan and actively mapping to the United Nations Sustainable Development Goals for Quality Education and Reduced Inequalities.

Date Published:2026