EPAM School of Digital Technologies (capstones)

Permanent URI for this collection

https://er.auk.edu.ua/handle/234907866/4

Browse

Now showing 1 - 19 of 19

DEVELOPMENT OF SOCIAL AND EVENT MANAGEMENT DISTRIBUTED APPLICATION
(2024) Kulomin, Dmytro
Purpose of this work is a creation of the product helping people and business to find each other, plan and organise activities and events. During the work on the problem different use-case were analysed and matched against existing solutions on the market. Work consists of observation of current state of industry, available products comparison, fit map creation and main use-cases identification. The result of the work is presented by the engineering solution in the form of main architectural decisions, storage and communication approaches analysis and the proof of concept implementation.
RECOMMENDATION SYSTEM FOR DATING PROJECTS
(2024) Mostovyi, Oleksandr
The main functionality for all dating projects is to elevate user experience through the provision of potential matches based on a variety of factors. This project is dedicated to evaluating and comparing the performance of diverse recommendation algorithms within the unique context of dating platforms. As a result an environment consisting of recommendation system and integration into search system of dating application is provided. The evaluation metrics include but are not limited to accuracy, precision, recall, and user satisfaction. By systematically testing these algorithms in controlled scenarios, the research seeks to identify specific performance strengths and weaknesses of each algorithm. Recommendation system is implemented using Python programming language and Flask framework and it uses dataset of profile from dating project. The research also explores the impact of diverse user behaviors and preferences on the recommendation algorithms, providing insights into the adaptability and robustness of each approach. Through a comprehensive analysis, this research aims to contribute valuable insights for dating projects seeking to enhance their recommendation systems. The findings will aid in the informed selection of recommendation algorithms tailored to the specific requirements and dynamics of dating platforms, ultimately improving user experience and satisfaction.
CUSTOM CRM SYSTEM FOR A LEGAL DEBT COLLECTION COMPANY
(2024) Bairaktar, Dmytro
A SUBSYSTEM OF THE INTEGRATION OF THE EDUCATIONAL PORTAL WITH THE LLM-BASE EDUCATIONAL CONTENT GENERATION SYSTEM
(2024) Ivchenko, Oleksandr
LLM models have created a new era of learning and teaching methodologies. This article examines the profound impact of LLM models on the educational environment. The study explores their integration in various educational settings and highlights their universal application and transformative potential. LLM models have the potential to bridge gaps in education, provide an alternative perspective, generate high-quality resources, and provide exclusive education for people with different learning needs. The study highlights the potential of combining technology and human experience to create an enriched learning ecosystem. The future of learning, shaped by artificial intelligence technologies, promises an improved educational experience.
APPLICATION OF OPENAI API FOR SEMANTIC CONTENT ANALYSIS IN INTELLIGENT E-LEARNING SYSTEMS
(2024) Rybak, Mykola
The thesis aims to describe the practical aspects of using OpenAI API in intelligent e-learning systems. The proposed work contains three main parts. The first part covers a background overview, theoretical aspects of a Large Language Model (LLM), transformer-based Natural Language Processing (NLP) architecture description, and use cases of different transformers. Introduces the term semantic unit and defines its main elements and connection with the capstone project. The second part concerns the rationale of the chosen model’s type for the capstone project and its description and implementation. This part covers requirements for the capstone project, C4 architecture diagrams, an overview of essential modules, libraries, and code examples of the implemented solution leveraging OpenAI API. The final part demonstrates further research on architecture improvements for better semantic content analysis and interaction with intelligent e-learning systems, including a combination of encoder-decoder and embedding models in the solution.
ExoCoDe: MODELING TRANSITS EVENTS VIA STATISTICAL AND MACHINE LEARNING TOOLS
(2024) Karakuts, Denys
This capstone project uses statistical and machine learning algorithms to detect exocomet transits in TESS telescope data. Exocomets, distinguished by their unique, asymmetric light curves, present a detection challenge due to their subtle signatures compared to planetary transits and light intensity. We develop a framework that integrates data preprocessing, feature extraction, visualizations, statistical methods, and machine learning regressors to characterize these transits efficiently. The project is built upon the existing progress of astrophysical research. It aims to enhance our understanding of exocometary activity, uncovering the potential of machine learning and statistical analysis in astronomical data interpretation.
COMPARISON OF ARCHITECTURAL PATTERNS WITHIN iOS APPLICATIONS
(2024) Skrypchenko, Mykyta
This work delves into the strategic selection of software architecture for iOS applications, underscoring the alignment of architectural decisions with specific project constraints and goals. Initial discussions centered around the challenges in using software metrics to compare various iOS architectures, leading to the proposal of a simplified framework aimed at aligning architectural choices with defined business objectives. The paper details the process of evaluating different architectural patterns — MVC, MVVM, VIPER, and TCA — considering these constraints and goals. This work contributes to the field by providing a practical example of how architectural decisions can be tailored to specific project constraints and goals, offering insights that can be valuable for software architects and developers working on similar Swift application projects.
HONEYCOMB MONOLITH: HEXAGONAL MODULAR PATTERN FOR AGILE MICROSERVICES EVOLUTION
(2024) Shablii, Taras
This thesis explores the architectural dilemma faced by startups and greenfield projects: choosing between monolithic and microservices structures. It addresses the gap in research on evolutionary monolithic architectures, introducing the Honeycomb Monolith pattern. This pattern combines Domain-Driven Design with Hexagonal Architecture to create modular monoliths poised for smooth transition to microservices. The effectiveness of the Honeycomb Monolith is demonstrated through the Opora application case study. This implementation validates the pattern viability, showing a seamless migration with minimal impact on the core domain logic. Challenges like model duplication and database management complexities are also identified, underscoring the need for strategic planning in architecture design. Concluding with future research directions, the thesis positions the Honeycomb Monolith as a viable solution for startups and an intermediary step for existing projects transitioning to microservices. This work contributes to the software architecture field, offering a novel solution that balances initial development efficiency with long-term scalability.
WEB SYSTEM WITH GRAPH UI FOR EXPLORATORY LEARNING
(2024) Tytenko, Andrii
This capstone project report details the development of a Graph User Interface application for Exploratory Learning. The report focuses first on the background overview of literature about graphs, exploratory learning, and existing web systems. Then it describes the development of a web-based graph visualization system where it highlights the selection of technologies like Next.js, TypeScript, and Tailwind CSS, and delves into the challenges encountered and UI features. The choice of specific library for graph visualization is emphasized for its performance and customization capabilities. The report also addresses user interaction, particularly the implementation of keyboard navigation to improve accessibility and user experience. It acknowledges the system's limitations, such as handling large data sets and extends into recommendations for future enhancements, including interface optimization and customization.
INTEROPERABLE WEB SYSTEM FOR GENERATION OF LEARNING RESOURCES ON THE BASE OF LLM AND PROMPT ENGINEERING
(2024) Kakun, Artem
The aim of the work is to create a system based on LLM and prompt engineering for generating educational materials. The proposed work consists of three main parts. Chapter 1 examines the progress of generative AI, such as Chat GPT, in education, focusing on its role in personalized learning and the importance of prompt engineering to maximize the effectiveness of AI usage. The proposed work outlines strategies for developing prompts to elicit accurate AI responses for educational purposes. Chapter 2 presents a detailed solution overview and architectural framework. The chapter outlines the business and architectural requirements, emphasizing the need for high scalability and portability. It also discusses the technology stack, the risks associated with relying on the OpenAI API, and the cost implications of the project. Chapter 3 summarises the work done on an AI-based educational material generation system. The chapter suggests areas for improvement, such as user authorisation, prompt updating mechanisms, skills analysis tools, and an educational chatbot. Overall, the successful integration of OpenAI models demonstrates the potential of the system in educational content generation.
DATA ANALYSIS AND LANGUAGE MODELS FOR D2C APPLICATIONS IN FOREIGN LANGUAGE LEARNING
(2024) Andriychuk, Sergiy
The landscape of foreign language learning has undergone a significant transformation in the digital age, particularly with the emergence of Direct-to-Consumer (D2C) applications. Relevance of the Topic At the forefront of educational technology, the integration of data analysis and language models in D2C language learning applications like WORDY represents a significant leap. Objective The primary objective of this study is to enhance the WORDY application for foreign language learning through the integration of advanced data analysis techniques and sophisticated language models. Structure of the Report This report offers a comprehensive analysis of the application of data analysis and language models in the WORDY application. It starts with a detailed literature review, leading into an exploration of the WORDY application's design. This design incorporates a thoughtful selection of technologies and frameworks to ensure optimal performance and user experience.
QUANTITATIVE RISKS ANALYSIS ON SOFTWARE PROJECTS USING THE MONTE CARLO METHOD
(2024) Oliyarnyk, Yuriy
Software projects are crucial for creativity, productivity, and organizational performance. However, their complexity can lead to risks and failures. Despite modern project management tools improving success rates, significant investment waste and project failures persist. In 2020, 11.4% of investment was lost due to poor project performance, and in 2021, 33% of IT projects failed, resulting in significant budget losses. As organizations enter 2022, challenges in big data, analytics, and AI projects are increasing, highlighting the need for a paradigm change in project management techniques. The Monte Carlo approach, a quantitative risk analysis technique, is gaining popularity to reduce uncertainty in software projects and provide more reliable estimations. The current implementation uses Jupyter Notebooks technology and decreases a steep learning curve for users.
WEB ACCESSIBILITY OPTIMIZATION IN FRONT-END DEVELOPMENT
(Manuscript, 2025-01-23) Starchenko, Anastasiia
This capstone project explores the critical importance of web accessibility in front-end development, emphasizing inclusivity for users with diverse abilities. With over 1 billion individuals globally experiencing some form of disability, this study focuses on creating a web accessibility analysis and optimization web app. The research identifies gaps in existing tools like Lighthouse, which often miss nuanced accessibility issues. Using WCAG standards, the app analyzes HTML code to detect violations, such as missing alt text, improper ARIA roles, and low contrast ratios. Featuring visualization, recommendations, and downloadable reports, the app aids developers in achieving WCAG AA compliance. The project underscores the value of inclusive digital design, offering practical implications for developers, businesses, and educators, while setting a foundation for future enhancements, including AI integration and internationalization.
A SOFTWARE SOLUTION FOR INTELLIGENT FAULT DETECTION IN WHEEL BEARINGS
(Manuscript, 2025-01-23) Glugovskiy, Glib
Wheel bearings are critical components in machinery and vehicles, and their failure can lead to costly downtime and safety risks. This project focuses on the research, development, and implementation of an intelligent software solution for fault detection in wheel bearings. By combining advanced signal processing techniques and machine learning, the solution enables early and accurate fault diagnosis efficiently for all users. The software incorporates Fast Fourier Transform (FFT) for feature extraction and Self-Organizing Maps (SOMs) for clustering and fault classification, achieving high accuracy in detecting bearing defects. This approach marks a significant advancement in the fields of machine maintenance and fault diagnostics by offering a more precise and efficient method for identifying defects in wheel bearings. A key aspect of the project involves the development and training of a SOM model. This model is specifically designed to analyze patterns in the cyclostationary signal data, enabling the accurate identification of potential bearing faults. The SOM model, along with the other signal processing and machine learning components, is integrated into a robust software application. Built using Django and React.js for the web interface and enhanced with Rust libraries for computational efficiency, this solution presents a novel integration of traditional web technologies with modern performance optimization techniques.
MULTIMODAL RETRIEVAL AUGMENTED GENERATION SYSTEM
(Manuscript, 2025-01-23) Yuvzhenko, Denys
This study presents an asynchronous, web-based Retrieval-Augmented Generation (RAG) system that integrates multimodal inputs (text, images, tables) to enhance information retrieval and generation in various contexts. The system is developed in Python and hosted on AWS, combining Chroma DB for vector storage and Anthropic’s Claude-3-Haiku LLM model, which is accessible via AWS Bedrock. By leveraging modern cloud capabilities, the solution scales efficiently and handles diverse data modalities in real time. Through systematic experiments, this project highlights the effectiveness of multimodal embedding techniques for refining retrieval accuracy and providing context-aware responses. The architecture’s modular design supports seamless feature integration, making it adaptable for different use cases such as customer support, educational tools, and content creation. Key findings emphasize the role of vector databases in dynamic information updates and confirm that large language models, when appropriately curated and grounded, can produce high-quality, relevant outputs
DEVELOPMENT OF ENHANCED MACHINE LEARNING METHODS FOR THE AUTOMATIC MODULATION CLASSIFICATION OF QAM SIGNALS
(Manuscript, 2025-06-09) Savenkov, Andrii
This study evaluates multiple machine learning approaches for Automatic Modulation Classification (AMC) of five digital modulation schemes: BPSK, QPSK, 8PSK, 8QAM, and 16QAM. The research implements and compares four classification methods: Gaussian Mixture Models, multinomial regression with polynomial features, nearest neighbor approach, and Convolutional Neural Networks. The system is integrated into a web service using Elixir/Phoenix framework and Python. Experimental results demonstrate high accuracy across all methods, with CNNs achieving 99% accuracy on simulated signals and consistent 99% accuracy on real signals from a CDM-625 modem. The study provides a practical solution for deployment in modern communication systems.
MACHINE LEARNING METHODS AND SOFTWARE TOOLS FOR PREDICTING CYCLICAL ECONOMIC PROCESSES IN A STOCK MARKET
(Manuscript, 2025-06-09) Saveliev, Myron
This capstone project focuses on forecasting cyclical economic trends in the stock market, using Apple Inc. (AAPL) stock data as a case study. Key research questions include: (1) How effective are machine learning models in forecasting market cycles? (2) Which models provide the most accurate predictions? Historical stock price data from 2010 to 2023 was collected from Yahoo Finance, and seven forecasting models were implemented: ARIMA, SARIMA, Gradient Boosting, Random Forest, LSTM, GRU, and Prophet. The research involved preprocessing the data, performing exploratory analysis, and evaluating the performance of each model using key metrics, including RMSE, MAPE, and R². Results demonstrated that ARIMA and SARIMA provided the most accurate forecasts, achieving an RMSE of 3.11, MAPE of 0.01, and R² of 0.98. Among deep learning models, LSTM outperformed GRU with an RMSE of 4.35 and R² of 0.96, effectively capturing non-linear dependencies. Gradient Boosting, Random Forest, and Prophet models underperformed, with higher RMSE and negative R² values, making them less suitable for this task. The study concludes that ARIMA and SARIMA are optimal for efficient and accurate forecasting, while LSTM is well-suited for complex, non-linear patterns. Future work could explore the incorporation of macroeconomic indicators, hybrid models, or alternative data sources, such as sentiment analysis, to enhance predictive accuracy. This research offers practical tips for utilizing machine learning in financial forecasting.
3D VISUALIZATION OF A ROOM USING A GROUND-BASED DRONE
(Manuscript, 2025-06-09) Shevchenko, Dmytro
This project presents the design and implementation of a low-cost, mobile 3D scanning system using the Intel RealSense D435i RGB-D camera mounted on a wheeled rover. The system performs indoor environment reconstruction by alternating between precise 360° panoramic scans and short forward movements, accumulating depth data at each waypoint. Combining IMU feedback, servo control, and depth sensing, it builds point clouds without high-end LiDAR. The result is a 3D model suitable for tasks like inspection, logistics, or demining support. The modular software stack, based on Open3D and Python, enables fast prototyping and flexible deployment in real-world environments
AUTOMATED QUIZ CREATION USING CHATGPT WITH INTEGRATION INTO THE CANVAS LMS
(Manuscript, 2025-06-09) Lazebnyi, Vitalii
This study explores the automation of quiz generation using Large Language Models (LLMs) like GPT-4, and their integration into Learning Management Systems (LMS) Canvas. It discusses the potential of LLMs to transform assessment creation by generating diverse question types aligned with learning objectives. The research also addresses ethical considerations, including data privacy and algorithmic bias, and highlights the importance of prompt engineering for effective AI-human interaction. The findings suggest that LLMs can significantly streamline quiz creation, saving educators time and enhancing student learning experiences while emphasizing the need for ongoing refinement and ethical oversight.

Browse

Browsing EPAM School of Digital Technologies (capstones) by Issue Date

Results Per Page

Sort Options