Bachelor of Science in Computer Science
Permanent URI for this collection
Browse
Recent Submissions
Item AI-Driven Personal Budgeting and Financial Literacy Enhancement for University Students in Uganda(Uganda Christian University, 2026-06-01) Oscar Odongkara; Precious Wakabi Namaganda; Joseph MukamaUganda’s university students face a critical financial management challenge. Despite mobile money reaching 33.8 million registered accounts and daily transaction values exceeding UGX 1.2 trillion, fewer than 25% of students aged 18–30 maintain any form of personal budget, and only 34% demonstrate basic financial literacy. Existing global applications such as Mint, YNAB, and Wallet are designed for structured banking environments and cannot parse the unstructured receipt images, nor do they recognise culturally relevant expenditure categories unique to Ugandan student life. This project presents UniGuard Expense Tracker, an Progressive Web Application (PWA) that integrates artificial intelligence, Receipt OCR parsing, predictive budgeting, and gamified financial education to improve personal financial management and literacy among university students in Uganda. The system implements a Receipt OCR Scanner pipeline with OCR and Random Forest classification to automatically categorise transactions into 18 locally meaningful categories. A hybrid Random Forest anomaly detection architecture optimised via Particle Swarm Optimization (PSO) provides accurate spending forecasts from limited historical data. A gamification engine grounded in Self-Determination Theory delivers points, badges, leaderboards, and daily challenges, while a financial literacy Gemini AI Companion provides context-aware financial coaching. Evaluation of the system produced strong results across all three objectives. The Receipt OCR scanner achieved an overall parse success rate of 96.4% on Ugandan mobile money receipts, while the Random Forest transaction classifier reached 89.2% accuracy across 18 culturally localised expenditure categories. The PSO-tuned forecasting model recorded a Mean Absolute Percentage Error (MAPE) of 8.9%, outperforming both ARIMA and GRU baselines. Usability testing with 25 Uganda Christian University students yielded a System Usability Scale (SUS) score of 83.4, placing the application in the “Excellent” usability category. An 8-week pilot with 97 students demonstrated a statistically significant 23.9 percentage point gain in overall financial literacy (p < 0.001), with the largest improvement recorded in budgeting confidence (+27.5%). The 28-day user retention rate reached 67.4%, confirming sustained engagement with the gamified learning framework. The total development cost remains within UGX 3,190,000, making the solution practically achievable in the Ugandan tertiary context.Item Image classification on a smart bin for waste segregation and analytics(Uganda Christian University, 2026-06-01) Arthur Shalom Yawe; Priscila Denise Muwanguzi; Nicole Mbabazi MwebembeziThis project addresses the problem of inefficient waste sorting methods currently in use, such as bins with well labelled compartments that rely on users to sort waste correctly. In addition, manual sorting by waste collectors at the point of collection is unsafe due to exposure to hazardous materials. There is also a lack of sufficient data to support effective waste management and decision-making by organizations such as Kampala Capital City Authority (KCCA) and National Environment Management Authority (NEMA). To address these challenges, we developed a smart bin with three compartments for paper, plastics, and general waste. The system allows a user to dispose of a single item of waste, after which the bin automatically sorts it into the correct compartment. At the same time, data is collected and sent to a dashboard for analysis of waste trends. The system works by using a camera to capture an image of the waste item. The image is then processed by a VGG16 deep learning model, which classifies the type of waste based on its visual features. The classification result is used to guide the sorting mechanism, and the data is transmitted to an application for further analysis. The backend is built with Node.js and Express, using MongoDB for data storage, while the frontend dashboard is developed with React and Tailwind CSS. The model was trained on a custom dataset of over 500 waste images collected at Uganda Christian University, supplemented by transfer learning from the VGG16 architecture pre trained on ImageNet. Testing demonstrated a classification accuracy exceeding 90% for the targeted waste categories. The integrated system successfully sorts waste into the correct compartments and provides real-time analytics on waste disposal patterns. This solution contributes to improved waste segregation at the source, reduces occupational hazards for waste workers, and provides data-driven insights for institutional waste management.Item ANI-LINK: AI-Powered cattle foot-and-mouth disease detection and veterinary care in Uganda(Uganda Christian University, 2026-06-04) David Buembo; Leon Matabi Kasingye; Teopista NajjumaLivestock production plays a critical role in rural livelihoods across Uganda, providing income, food security, and economic stability for smallholder farmers. However, the sector continues to face significant challenges due to delayed disease detection, limited access to veterinary services, and fragmented livestock health management systems. These challenges contribute to substantial productivity losses, particularly from diseases such as Foot-and-Mouth Disease (FMD). This study presents AniLink, an artificial intelligence (AI)-enabled livestock health platform designed to support early disease detection, improve health record management, and enhance access to veterinary services. The system integrates a multi-task deep learning model based on MobileNetV3 for simultaneous cattle verification and FMD detection, alongside digital health records and a mobile- based service interface. A Design Science Research (DSR) approach was adopted to guide system development and evaluation. The model was trained on a combination of publicly available datasets and locally collected images from Gomba District, Uganda. Evaluation was conducted using classification metrics, system performance measures, and user-centred usability testing. Results demonstrate strong model performance, achieving high recall for both cattle detection and FMD classification, along with efficient real-time processing suitable for mobile deployment. Usability evaluation indicates that the platform is accessible to users with varying levels of digital literacy, with positive feedback on ease of use and clarity of outputs. The findings suggest that integrated AI-driven livestock health systems can significantly improve disease detection, decision-making, and access to veterinary support in resource-constrained environments. The study contributes a practical and scalable framework for applying artificial intelligence in agricultural health management.Item Glucosense Ecosystem: An Intelligent Clinical Insulin Decision Support and Nutrition Platform for Diabetes Care.(Uganda Christian University, 2026-06-04) Joy Abaho; Godfrey Mucunguzi; Bachawa WangoloType 1 diabetes is one of the most surging health condition with an approximate number of 369,100 adult cases in Uganda by 2024. There is still a lack of adaptive and personalized tools necessary to manage the disease. The GlucoSense ecosystem focuses on an integrated, full-stack with machine learning to predict insulin intakes, a nutritional system that offers meal recommendations, a meal chatbot, and glucose-guided foods for diabetic patients. The project was built on a three-tier architecture with the frontend layer developed in ReactVite for both the Clinical portal and the Meal plan User Interface, the backend layer follows a FastAPI design and the database layer was built in SQLite and SQLAlchemy. The results show that the Linear Regression model achieved the lowest error values in predicting insulin dosage with a Mean Absolute Error of 2.98, Mean Squared Error of 11.524 and a Root Mean Squared Error of 3.3947. These results proved that linear Regression was the most stable model and produced fewer dangerous miscalculations compared to random forest, gradient boosting and xgboost. However, all the models showed very low negative R2 values which would probably have occurred because of feature limitations since most features might not have been fully captured.Item IoT-enabled smart waste monitoring and predictive route optimization system(Uganda Christian University, 2026-06-01) Itungo Agaba; Alvin Rubagumya; Mark Calvin ObbaRapid urbanization in Uganda has led to a significant increase in solid waste generation, while existing collection systems remain largely manual, inefficient and reactive. Municipal waste collection typically relies on fixed schedules without real-time visibility into bin fill levels, resulting in overflow, increased operational costs and environmental risks. This project presents an IoT-enabled Smart Waste Monitoring and Predictive Route Optimization System designed to address these challenges. The system integrates smart bin sensors, real-time telemetry, predictive analytics and route optimization algorithms to enable a data- driven waste collection process. The proposed solution utilizes ultrasonic sensors and Arduino UNO R4 microcontrollers to monitor bin fill levels, combined with machine learning models to forecast waste accumulation patterns. A routing optimization engine based on Vehicle Routing Problem (VRP) techniques is implemented to generate efficient collection routes. Results from simulation and system testing demonstrated that the proposed system improved operational efficiency through real-time telemetry monitoring and predictive routing. The forecasting models were able to identify high-priority bins before overflow occurred, while the route optimization engine reduced unnecessary collection trips and overall travel distance. System testing further confirmed reliable telemetry transmission, responsive dashboard visualization and effective integration between IoT devices, backend services and machine learning components. The system in general demonstrates the potential to reduce operational costs, prevent bin overflow and improve urban sanitation. The project contributes a scalable and cost-effective solution tailored for Ugandan municipalities, aligning with Sustainable Development Goals (SDG 11 and SDG 12) on sustainable cities and responsible resource management.Item Bako Analytics: a localized deep learning framework for personalized basketball biometrics and team tactical analysis(Uganda Christian University, 2026-05-28) Norbert Okidi; Anna Akumu; Zahara NankyaThe advancement of Artificial Intelligence (AI) in sports has largely bypassed low-resource environments in Sub-Saharan Africa due to the prohibitive cost of commercial hardware and the lack of representative datasets for African athletes. This project presents BAKO Analytics, Africa’s first localised multi-model deep learning framework designed for personalised basketball biometrics and tactical diagnostics. The system addresses the gap in accessible sports science in Uganda by implementing a decou- pled client-server architecture comprising a FastAPI-powered backend and a React/Vite web interface. The core innovation lies in a six-model AI pipeline using the YOLO (You Only Look Once) architecture for object detection and pose estimation. Specifically, the framework includes dedicated models for player and ball detection, court keypoint mapping (homography), and a custom biometric engine that decomposes basketball shooting form into four critical phases (DIP, SET, RELEASE, and FINISH) using 17-keypoint skeletal data. Validated through a case study with the UCU Cannons basketball team, BAKO Analytics enables elite-level performance analysis using nothing more than standard smartphone footage and modest computing hardware. Technical evaluation demonstrates high precision in action recognition and spatial tracking, even under the variable lighting conditions of outdoor courts common in the Ugandan context. Beyond its technical contributions, the project provides a scalable and affordable blueprint for indigenous sports analytics, bridging the technological divide and fostering talent development through data-informed coaching. Keywords: Basketball Analytics, Computer Vision, Deep Learning, Pose Estimation, Biometrics, Low-Resource AI, Sports ScienceItem Automated coffee bean quality sorting using machine learning and multi-sensor integration: A low-cost solution for Ugandan smallholder farmers(Uganda Christian University, 2026-06-03) Rebecca Alinda; Benjamin Mutumba Mubeezi; Isaac Nickson NzirigaUganda’s coffee sector, which contributes 20–30% of the country’s foreign exchange earnings, faces persistent quality challenges rooted in manual sorting practices. Ninety percent of Uganda’s coffee is produced by smallholder farmers who lack access to affordable sorting technology. Manual sorting is labour-intensive, inconsistent, and prone to human error rates of 20–25% after prolonged operation, limiting farmers’ access to specialty markets that pay 40–60% premium prices. This project presents the design, development, and evaluation of an Automated Coffee Bean Quality Sorting System, which is a low-cost (UGX 450,000–562,500) solution combining multi-sensor integration, machine learning, and embedded systems. The system employs a Raspberry Pi 4 as the central processing unit, a TCS3200 colour sensor for RGB-based bean assessment, a Camera Module 2 for visual inspection, and an Arduino Uno for motor control via an L298N driver. Beans are transported on a conveyor belt and assessed at sequential sensing stations before a servo motor diverts defective beans. The machine learning pipeline employs a dual-model fusion strategy: a Decision Tree classifier trained on colour sensor readings achieves approximately 98% accuracy, while a MobileNetV2 CNN model processed via TensorFlow Lite provides visual classification. A conservative fusion strategy defaults to rejection when model outputs conflict, ensuring food safety. The system achieves a throughput of 1–2 beans per second with inference latency below 50 ms. Key findings include: (1) the Decision Tree outperforms the CNN in the current deployment due to a domain gap between training data and Ugandan Robusta beans under real field conditions; (2) the conservative reject-on-conflict strategy provides a defensible default for food quality systems; (3) the total cost represents a 97–98% reduction versus commercial alternatives costing UGX 18,750,000–56,250,000. The system demonstrates technical feasibility and strong potential for adoption among smallholder cooperatives across East Africa. Keywords: coffee bean sorting, machine learning, Decision Tree, MobileNetV2, TensorFlow Lite, Raspberry Pi, multi-sensor integration, smallholder agriculture, Uganda, precision agriculture.Item A hybrid machine learning network intrusion detection system for Small and Medium Enterprises (SMEs)(Uganda Christian University, 2026-05-25) Absolom Orianga; Mark Travis Lufene; Allan Kagimu SsebattaCybersecurity remains a critical challenge for Small and Medium Enterprises (SMEs) in Uganda, where cybercrime increased by approximately 93.5% in 2024, causing losses of approximately UGX 72 billion. Approximately 40% of Ugandan SMEs have experienced cyberattacks, yet most continue to rely on basic signature-based tools that are incapable of detecting sophisticated or zero-day attacks. This project presents Vanguard-NIDS, a hybrid machine learning-based Network Intrusion Detection System designed for real-time intrusion monitoring in SME environments. The system captures live network packets using Scapy, extracts flow-level and statistical traffic features, and analyses them through a three-pronged detection pipeline comprising signature-based pattern matching, supervised machine learning via an ensemble of Random Forest, SVM, XGBoost, and LightGBM classifiers, and unsupervised anomaly detection via Isolation Forest, One-Class SVM, and Autoencoder models. A result fusion engine combines these predictions into a unified threat score. Models were trained and validated on the CICIDS2017, NSL-KDD, and UNSW-NB15 benchmark datasets. The Random Forest classifier achieved approximately 95% detection accuracy with a false positive rate of approximately 3%. A real-time React-based dashboard delivers live traffic monitoring, alert management, and system metrics via WebSocket communication, with an average end-to-end detection latency of under 25 milliseconds.Item Digital performance tracking for amateur football(Uganda Christian University, 2026-06-04) Gareth Neville Kisuze; Andrew Ogwang; Derrick Elvan KatendeGrassroots football draws millions of participants worldwide, yet most players still record performance through scattered channels such as paper notes, group chats, and one o! social posts. Records are hard to compare, easy to lose, and rarely support any real analysis. Tools built for elite clubs depend on GPS feeds, video pipelines, and specialist sta! that community leagues simply do not have. This report presents FootyStats, a mobile platform that lets amateur players and tournament organisers log match events during play, store them in a shared database, and view leaderboards, player profiles, and simple performance trends. The work combines user centred design with a lightweight client server setup and automated aggregation so that useful summaries can be drawn from ordinary event data rather than costly tracking hardware. Prototype testing showed that structured, low friction capture cuts down the fragmentation seen with informal methods, and that even basic match records can support meaningful summaries for grassroots users. The study offers both a practical path toward deployable amateur sports informatics and evidence that analytics can be delivered under the resource limits typical of community football.Item A Yolo-based robotic system for waste detection and Collection(Uganda Christian University, 2026-06-02) Isaac Nabasa; Apophia Atwijukire; Leticia LackiaWaste management remains a critical challenge in Kampala where approximately 63% of daily generated waste goes uncollected, leading to environmental pollution, blocked drainage systems, flooding and public health risks. Traditional manual collection methods are inefficient, labor intensive and inadequate to handle the growing volume of lightweight litter such as paper,npolythene bags, and plastic bottles, especially in busy public spaces. This project developed ECO-BOT, a low-cost autonomous robotic system that integrates computer vision and robotics to address these challenges. Following the engineering design method ology, the robot was built using the Hiwonder mobile platform, Raspberry Pi 5, a lightweight YOLOv8n object detection model, and a 5-DOF robotic arm. The system was designed to detect, navigate towards, pick up and store lightweight waste items. Individual components were developed and tested separately before full system integration. The complete prototype was evaluated in outdoor environments on surfaces including tarmac, short grass and light gravel and also inside Uganda Christian University Dining hall which has cemented ground and some tables. Results demonstrated promising performance with 87% detection accuracy, 78% collection success rate, and 84% navigation success across tests, achieving an average collection time of 13.5 seconds per object. While limitations such as occasional processing delays and challenges with certain waste types were observed, ECOBOT successfully shows that affordable YOLO-based robotics can provide a practical, scalable solution for improving waste collection in resource limited urban settings like Kampala. This work contributes to smarter, technology driven waste management practices in Uganda.Item Harmony Guardian: A wearable multi-sensor assistive system for stress and behavioural monitoring in individuals with neurodevelopmental disorders(Uganda Christian University, 2026-06-06) Wangobi K Nicholas; Mawejje J Paul; Asingura K PhilipNeurodevelopmental disorders (NDDs), such as Autism Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD), often impair an individual’s ability to regulate emotions, process sensory stimuli, and communicate distress effectively. These challenges can result in behavioural escalations, including meltdowns and shutdowns, which are commonly preceded by measurable physiological changes. In low-resource settings such as Uganda, caregivers largely depend on subjective observation due to the lack of affordable, real-time monitoring systems, leading to delayed interventions and reduced quality of care. This project addresses this problem through the development of Harmony Guardian, a personalised, lightweight, and sensory-friendly wearable monitoring system designed for individuals with neurodevelopmental disorders. The system integrates multiple physiological sensors, including heart rate variability (HRV), galvanic skin response (GSR), and motion sensors, to continuously monitor stress-related indicators and behavioural patterns. Embedded processing and real-time data analysis techniques are used to identify early signs of stress and behavioural dysregulation, after which alerts are transmitted to caregivers for timely intervention. The proposed solution was designed with emphasis on affordability, usability, portability, and suitability for under-resourced environments. System evaluation demonstrated that the wearable device was capable of reliably capturing physiological changes associated with stress responses and providing timely notifications to caregivers. The results indicate that the system can improve early intervention, enhance caregiving efficiency, and contribute to better behavioural and health outcomes for individuals with neurodevelopmental disorders. Overall, the project demonstrates the potential of accessible wearable technologies in supporting inclusive and proactive healthcare for people with special needs.Item AquaGuardian: An IoT-driven real time water quality monitoring and machine learning prediction for aquaculture systems in Uganda(Uganda Christian University, 2026-06-08) Calvin Tendo; Ronald Austine Ezamamti; Emmanuel KisaUganda’s aquaculture sector faces a critical productivity crisis, with national fish production declining by 27.8% and export revenue falling 21.9% between 2023 and 2024. This decline is largely attributed to inadequate real-time water quality monitoring among smallholder farmers, who continue to rely on manual testing methods that fail to detect sudden changes in critical parameters. This project presents AquaGuardian, an IoT-based smart fish pond monitoring and management system designed specifically for the Ugandan context. The system integrates a Raspberry Pi 4B controller with an integrated RS485 NPK sensor probe, a digital turbidity sensor and a GSM-based SMS alert module to enable continuous, real-time environmental surveillance. The architecture comprises five functional layers: Sensing, Processing, Actuation, Communication and Cloud Logging. A Random Forest machine learning model, trained on 3,887 samples from the Aquaculture Water Quality Dataset (AWD), classifies pond water quality into three categories: Excellent, Good and Poor. By utilizing biologically-informed feature engineering derived from NPK, pH and temperature data, the model achieved an 85.5% theoretical accuracy and 100% precision on the critical ”Poor” water quality class. Field testing validated the system’s ability to deliver real-time sensor visualization via a cloudhosted Flask dashboard deployed on Render and Neon.tech. The system successfully activated automated aeration and circulation pumps within five seconds of detecting a critical threshold and dispatched SMS alerts to farmers within thirty seconds. However, the evaluation also highlighted the limitations of using Nitrogen-based surrogate substitution for Nitrite inputs, a necessary compromise due to local hardware market constraints. This project demonstrates how the integration of IoT, machine learning and cloud technologies can bridge the monitoring gap in smallholder aquaculture,while providing a transparent assessment of the hardware-software alignment challenges encountered in resource-constrained environments.Item Tomato Doctor: An AI Advisory System for tomato leaf disease detection and retrieval-augmented agronomic guidance for smallholder tomato production(Uganda Christian University, 2026-06-11) Newton Anguyi; Timothy OpifudriraSmallholder tomato farmers require timely, practical, and localized guidance for tomato crop management decisions, yet many lack fast access to expert support when Tomato Early Blight (Alternaria solani) or Tomato Late Blight (Phytophthora infestans) symptoms first appear on Solanum lycopersicum. In this project, we built a full-stack Smart AI Advisory System for Smallholder Farmers, delivered to end users as the Tomato Doctor mobile application. The prototype integrates three tightly-coupled capabilities into a single workflow: (1) AI-based tomato leaf disease diagnosis from images, (2) a contextual advisory chatbot that turns a tomato diagnosis into actionable management guidance, and (3) optional field-feedback capture to support future dataset improvement and retraining. The system is intentionally narrowed to three classes only: Tomato Early Blight, Tomato Late Blight, and Healthy (Tomato Leaf). The disease diagnosis pipeline is implemented with transfer learning using MobileNetV2 [1], trained on a 3-class subset of PlantVillage [2] and then fine-tuned using field images to improve real-world robustness. The primary deployed model is the field fine-tuned checkpoint mobilenetv2_field_3class_fieldda ta.pth. Inference includes uncertainty gating to handle images that are not tomato leaves or are unclear; instead of forcing a wrong class, the API returns an “unsupported/unclear” response. Robustness is further improved by lightweight test-time augmentation (original + horizontal flip) and careful image preprocessing that preserves aspect ratio and corrects EXIF orientation. Explainability (Grad-CAM) [3] is generated for Early Blight and Late Blight predictions only, and severity estimation (lesion coverage %, lesion count, and severity stage) is applied to Early Blight detections to support urgency-aware recommendations. The advisory feature is implemented as an authenticated Django REST API that combines a retrieval-augmented generation (RAG) pipeline (sentence-transformers [4] with ChromaDB [5]), disease-context prompting using optional fields disease_name and detection_context, and a rule-based fallback for greetings and very short messages. Field evaluation on 89 held-out real photos shows the impact of fine-tuning: the base model achieved approximately 28% accuracy on field images, while the fine-tuned 3-class model achieved 85.4% accuracy with balanced precision/recall. The final outcome is a v1.0 proof-of-concept that integrates tomato foliar disease detection (Early Blight, Late Blight, Healthy), explainability, and grounded advisory guidance in a single mobile-first workflow, with clear limitations for real-world use.Item A hybrid intent-driven and retrieval-augmented conversational framework for reliable University information access(Uganda Christian University, 2026) Nicole Mary Johnson; Humphery Mubiru; Ethan ArikoStudents at Uganda Christian University frequently struggle to access timely and accurate institutional information because relevant details are scattered across notice boards, social media groups, and outdated web pages. This report presents the design, implementation, and evaluation of a hybrid multi-channel chatbot that merges intent-based routing, keyword retrieval, vector-based semantic search, and large-language-model generation to deliver context-grounded answers drawn from seventeen verified university documents. The system was deployed as both a web application and a WhatsApp integration, offering multilingual support in seven languages through the Sunbird translation service. Functional testing across representative query categories showed that the hybrid retrieval strategy returned relevant context for the majority of test queries, while a confidence-aware fallback mechanism directed users to appropriate university offices when the knowledge base could not provide a reliable answer. The findings indicate that combining complementary retrieval methods with grounded generation is a viable approach for building dependable information assistants in resource-constrained university settings.Item AI Powered Computer Vision-Based Framework For Real-Time Road User Classification And Intelligent Traffic Signal Control(Uganda Christian University, 2026-06-04) Rachel Mbeiza Isooba; Promise Pierre Mokili; Joshua Owor GenoTraffic congestion is a leading cause of economic and productivity loss in Uganda. The Greater Kampala Metropolitan Area carries roughly half of the country’s registered vehicles, and peak-hour speeds on central corridors drop to as low as 11 km/h. At the same time, more than 70% of road traffic fatalities in the country fall on vulnerable road users, that is pedestrians, cyclists and boda-boda motorcyclists. Most of Kampala’s signalised junctions still operate on fixed-time plans that cannot respond to real demand, and none of them offers cyclists a dedicated signal phase. This report describes RoadWise, a group final-year project that builds a low-cost, AI-powered traffic management framework for classifying road users in real time and controlling intersection signals dynamically. The system is organised in four layers: a sensing layer made of USB cameras and IoT sensors, an intelligence layer that runs a YOLOv8 detector and a Firebase cloud backend, an action layer that drives smart traffic lights, and an interface layer made of a traffic officer dashboard and a road user mobile web app. A custom scale model traffic setup was built, covering both three-way and four-way junction layouts, so that the full detection to actuation loop could be tested under controlled conditions.The work was grounded in a PRISMA compliant systematic literature review that screened 4,419 records and retained 12 primary studies. The review confirmed that, although YOLO based vision and adaptive signal control are individually mature, no reviewed system offers a dedicated cyclist priority phase in mixed traffic. RoadWise closes that gap through what we call a blue light phase, which is activated whenever cyclist presence at a junction passes a configurable threshold. On the prototype, the detector reached 100% vehicle detection and 85% cyclist detection accuracy under varied lighting, with a best validation mAP@0.5 of 0.977 on the miniature model dataset. The adaptive controller reduced junction waiting times by up to 40% compared with a fixed time baseline, and the web to hardware synchronisation protocol achieved 100% state consistency over serial acknowledgement feedback. The same controller was shown to run simultaneously across three-way and four-way configurations, which supports scalability toward real junctions. The report also discusses the limits of the work, in particular the scale model-to-real domain gap, the narrow class schema, and the remaining steps needed before a pilot deployment on a real KCCA junction can be attempted.Item AI-Powered Robotics for Post-Disaster Survivor Detection and Rescue Path Mapping(Uganda Christian University, 2025-05-06) Diana NansubugaNatural and man-made disasters often highlight the weaknesses in traditional emergency response efforts, especially in countries with limited resources like Uganda. Challenges such as slow response times, difficulty reaching affected areas, and a lack of real-time information frequently result in lost lives and misused resources. This study introduces a robotic dog system enhanced by artificial intelligence, designed to improve the effectiveness of search and rescue missions following disasters. The system incorporates real-time human detection using AI, SLAM for creating environmental maps, and the A* algorithm to plan efficient rescue routes. The robot is powered by a Raspberry Pi and managed through a Flask-based web platform. It includes multiple sensors, such as GPS, ultrasonic detectors, a night vision camera, and directional microphones, allowing it to navigate independently, avoid obstacles, and communicate wirelessly with remote rescue teams. Testing the system in a controlled, disaster-like setting confirmed its ability to locate victims, generate accurate maps, and determine safe paths with high reliability. The results suggest that. This low-cost, locally adaptable solution could play a vital role in speeding up rescue efforts and minimising risks to human responders. By addressing key limitations in current practices, the project adds valuable insights to the field of disaster robotics, particularly In settings where advanced tools are not readily available.Item Gandabert: Transfer Learning With Mbert for Luganda News Classification(Uganda Christian University, 2025-05) Seth MbashaLuganda, spoken by over 21 million Ugandans, is significantly under‐resourced in Natural Language Processing (NLP), lacking effective tools like news classifiers. This gap hinders digital information access and contributes to the digital language divide. This research project addressed this challenge by developing GandaBERT, a model for Luganda news classification. The methodology involved fine‐tuning the multilingual BERT (mBERT) model on a novel multi‐source dataset comprising 2,609 native, translated, and synthetic Luganda news articles across five categories (Politics, Business, Sports, Health, Religion). Evaluation on a held‐out test set showed GandaBERT achieved an overall accuracy of 85.7%. While demonstrating strong performance in certain categories like Politics, challenges and variations across topics were observed, partly linked to overfitting during training. This study confirms the viability of applying transfer learning with mBERT for practical Luganda NLP tasks, provides a valuable classification tool, and contributes towards enhancing digital resources for this low‐resource language.Item An Embedded and Machine Learning Based Early Flood Monitoring and Warning System, the Case of River Manafwa(Uganda Christian University, 2025-05-06) Daniel Lukyamuzi WavamunnoFlooding remains a serious threat in many parts of Uganda, especially in regions with limited access to early warning systems. This project introduces a practical solution that combines embedded hardware and machine learning to monitor and predict flood events in real time. Using a flow sensor and an ultrasonic sensor connected to an ESP32 device, the system captures data on water movement and levels. These readings are automatically logged to Google Sheets, allowing for easy data management and access. A backend built with FastAPI processes this information, using a trained Random Forest algorithm to forecast potential flood risks. The results, along with past records, are displayed on an interactive dashboard developed in React. By merging simple electronics with predictive analytics, the system provides an affordable and adaptable tool to support timely flood response efforts in vulnerable areas.Item AI Image-based System for Lumpy Skin Disease Detection in Cattle(Uganda Christian University, 2025) Amos MugabiLumpy Skin Disease (LSD) remains a significant threat to cattle health across Uganda, with conventional disease detection methods being slow, centralized, and reliant on clinical expertise that is often unavailable in field settings. This project proposes an innovative solution through an AI-powered, image-based detection system capable of identifying LSD from cattle images. The system employs a two-stage deep learning architecture: a YOLOv8 object detection model locates individual cattle within images, followed by a convolutional neural network (CNN) that classifies each animal as either healthy or infected based on visible skin lesions. Trained on a diverse dataset of annotated cattle images, the integrated model achieved a high detection precision and classification accuracy, demonstrating strong reliability in recognizing signs of LSD. Furthermore, the system offers real-time feedback via an interactive web interface, enabling farmers and veterinary personnel to quickly assess cattle health with images. This approach not only enhances detection and control measures but also sets the stage for broader adoption of AI in livestock health management within low-resource environments. The system’s design aligns with global goals of smart agriculture, offering a scalable tool that supports both food security and disease resilience.Item SMART-AG : A Precision Agriculture AI-Powered Edge Computing System(Uganda Christian University, 2025-05-06) Totit Kabuya BushenyulaFood insecurity is a critical issue in Africa, with over 296 million people affected by hunger. Despite efforts to increase food production, food loss remains a major contributor to this crisis, particularly in low-income countries where inefficiencies in farming and post-harvest handling are common. While 42% of Africa's workforce is employed in agriculture, there is a low adoption of modern agricultural technologies, primarily due to lack of internet access, technical skills, and high costs. This results in many farmers continuing to rely on traditional methods, which limits their productivity and exacerbates food insecurity. Existing solutions to improve farming practices are often too complex or require constant internet access, leaving many farmers unable to benefit from them. Therefore, a practical, affordable, and internet-independent solution is needed to help farmers increase yields and reduce food losses. This project proposes SMART-AG, an AI-powered edge computing system that provides actionable farming advice via SMS without requiring internet access. SMART-AG aims to empower farmers with insights on soil health, crop selection, and nutrient management, improving productivity and contributing to food security in Africa.