http://hdl.handle.net/123456789/17504 2026-11-14T10:20:52Z 2026-11-14T10:20:52Z 03-134222-003, ABDUL MUTLIB ALI http://hdl.handle.net/123456789/21737 2026-10-06T05:34:01Z 2026-06-01T00:00:00Z

SKYLYTICS AN AI POWERED FLIGHT DELAY PREDICTION SYSTEM 03-134222-003, ABDUL MUTLIB ALI Flight delays cost the aviation industry over $28 billion annually, yet most systems remain reactive, and existing research often fails to translate predictive models into real-world operational tools. Skylytics addresses this by providing a full-stack, proactive flight delay prediction system designed for both operational managers and passengers. It uses a hybrid stacking ensemble combining XGBoost and Bidirectional LSTM models, integrated through a Logistic Regression meta-learner, leveraging their complementary strengths. The system is trained on around 14 million flight records (2024–2025), enriched with weather data from Open-Meteo and live air traffic data from OpenSky Network. Multiple models were evaluated under a leakage-free setup, with XGBoost achieving a ROC-AUC of 0.7935, while the BiLSTM contributed significantly to the ensemble. The system is deployed via a FastAPI backend, a Next.js operations dashboard, and a SwiftUI iOS app for passenger tracking. It also includes. SHAP-based interpretability and an AI assistant for natural-language insights. Overall, Skylytics bridges the gap between academic research and production systems, enabling proactive decision-making and improving reliability for both airlines and passengers. DR. NADEEM SARWAR

2026-06-01T00:00:00Z 03-135222-009, Arooj Ishtiaq http://hdl.handle.net/123456789/21740 2026-10-06T05:38:13Z 2026-06-01T00:00:00Z

Bachelor of Science in Information Technology 03-135222-009, Arooj Ishtiaq The goal of this project is to create a complete digital platform called ‘Skillink' that will access a single ecosystem for connecting students, mentors and recruiters. People typically use several individual platforms to find jobs, find mentors, and acquire new skills in the digital world. Skillink overcomes this by incorporating all these features into a single, easy-to-use mobile app. This report dives deep into the development process of a complete mobile app developed with the Flutter platform and connected to a Supabase database.This system is capable of providing authentication, role based access control, real time database. Administrative controls and manipulation of user management. Additionally, the project includes a data analytics aspect which uses Microsoft Power BI to create customer engagement, app trends visualization. Mentoring and income opportunities Dashboard offers useful information for making decisions and growing platforms. The application is developed following a structured approach: requirement Analysis, design of systems, database modeling, implementation and testing. User management, mentorship system, job application tracking and admin dashboard. Among the main modules are to monitor and control platform activities. The system has been tested by various users and implemented successfully. essential features of login, job applications, mentorship requests and the administrative aspects. The findings suggest that the proposed solution offers an improvement in the use of the system and in its efficiency in relation to the current disorganized platforms. Dr. Junaid Nasir

2026-06-01T00:00:00Z 03-134241-036, Maaz Imran http://hdl.handle.net/123456789/21720 2026-10-05T09:26:48Z 2026-05-01T00:00:00Z

Fun Learn AI 03-134241-036, Maaz Imran FunLearn AI is an intelligent, AI-powered educational platform designed to provide engaging and adaptive learning experiences for children aged 3 to 12 years. The platform represents a holistic solution to early childhood digital education, addressing key deficiencies identified in existing educational technology products including the lack of voice-first interaction for pre-literate learners, the absence of genuine adaptive intelligence, and the commercial barriers that prevent equitable access. FunLearn AI was motivated by the significant digital divide in educational technology access across Pakistan and similar developing nations. Most available platforms require reading proficiency, charge subscription fees, and are designed for Western learners with no consideration of regional linguistic contexts. They also lack genuine AI features that adapt to individual learning trajectories. FunLearn AI was conceived to address these gaps by providing a freely deployable, voice-driven, AI-adaptive solution aligned with SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities). The platform leverages AI voice recognition via Web Speech API, adaptive difficulty via scikit-learn machine learning, and a comprehensive multi-role management system. The architecture comprises a Django REST Framework backend, a React.js frontend with Framer Motion animations, MongoDB Atlas cloud database, and JWT authentication. Nine games span three age groups Little Explorers (3-6), Junior Learners (6-9), and Super Scholars (9-12), each incorporating voice recognition that allows pre-literate children to interact naturally. Four portals serve students, teachers, parents, and administrators with role-specific analytics, class management, badge gamification, and progress reporting. Security employs OTP email verification, JWT tokens, admin secret key, and password strength enforcement. The DevOps infrastructure includes Docker, GitHub Actions CI/CD, Render (backend), and Vercel (frontend). Performance evaluation confirms that the platform meets all defined benchmarks: login API average response time of 0.8 seconds, game score submission latency of 1.1 seconds, AI progress report generation in 2.3 seconds, and Vercel CDN frontend load time of 1.6 seconds (see Table 6.3 for full performance test results). Voice recognition accuracy across phonetics-matched child speech achieved a 30-40 percentage point improvement over single-transcript baseline evaluation. Testing confirms successful voice recognition, OTP delivery, stage unlocking, and stable cloud deployment across all environments. Dr. Nadeem Sarwar

2026-05-01T00:00:00Z 03-134222-065, Muhammad Huzaifa http://hdl.handle.net/123456789/21722 2026-10-05T09:33:16Z 2026-05-01T00:00:00Z

SafeVision: AI-Powered Surveillance System 03-134222-065, Muhammad Huzaifa Conventional closed-circuit television (CCTV) surveillance infrastructure operates almost entirely in a reactive mode: cameras record footage and human operators decide, after the fact, whether anything significant occurred. This passive approach carries well-documented shortcomings. Research has established that human attention during continuous screen monitoring drops sharply after twenty minutes [14], making real-time threat detection from human observation alone an unreliable process at scale. SafeVision addresses this gap with an AI-powered, two-stage video anomaly detection system built on modern deep learning. Stage 1 screens every video clip for normal or anomalous activity. When an anomaly is detected, Stage 2 immediately classifies the type of incident into one of three super-categories: Fighting/Violence, Armed Threat, or Property Crime. Results are delivered through a polished, operator-facing web dashboard built with Next.js [18] and FastAPI [17]. The core of the system is a feature-extraction and temporal-reasoning pipeline. CLIP ViT-B/16 [4], a Vision Transformer pre-trained on 400 million image-text pairs, encodes each video frame into a 512-dimensional semantic embedding. A lightweight TemporalTransformer, a two-layer encoder-only transformer with sinusoidal positional encoding [6] processes sequences of sixteen frame embeddings to produce a single clip-level classification. Training used a unified dataset combining the UCF-Crime benchmark [1] (1,900 surveillance videos across 13 crime categories) and the Smart-City CCTV Violence Detection Dataset (SCVD) [2]. Stage 1 achieved a validation macro F1 score of 0.8950, supported by a macro precision of 0.90 and a recall of 0.90. Stage 2 achieved a macro F1 score of 0.7565, with a macro precision of 0.76 and a recall of 0.75 on three-class crime classification. The entire inference stack requires under 750 MB of VRAM, making it deployable on commodity hardware. Limitations include the absence of real-time RTSP streaming and frame-level temporal localisation, both identified as the highest-priority directions for future work. Dr. Muhammad Saqib Sohail

2026-05-01T00:00:00Z