]]> AND MACHINE LEARNING
]]> Several models were tested, including Logistic Regression, Support Vector Machines, GRUs, and transformer-based models like RoBERTa and DeBERTa. The best performance came from RoBERTa with label smoothing and DeBERTa-v3, both reaching 98% accuracy. While these advanced models were the most accurate, simpler models like GRU were still competitive and easier to interpret.
The thesis also examined linguistic differences between real and AI-generated reviews. Real reviews were shorter, used more personal and emotional language, while AI-generated ones were longer, more structured, and often overused formal or generic phrases.
A working browser extension was built as part of the project. It allows users to analyze reviews directly on websites and see predictions with basic explanations. Although the tool works well, there are still limitations, such as handling newer AI models and providing clearer feedback for non-technical users.
]]> The system was implemented using a modular, role-based structure that enables patients to search for doctors, view profiles, book or cancel appointments, and make payments online. Doctors can manage their schedules and track appointments, while administrators are provided with tools to oversee users and system activity. Methods included designing user-centered interfaces, applying secure authentication mechanisms, and testing the system through automated end-to-end simulations.
]]> ]]> This project aims to design and implement an automated antenna rotator system capable of tracking satellites and facilitating radio communication. The system integrates a custom-built hardware platform with a software interface to: (1) automatically adjust the antennas direction based on the position of the ground station and a chosen satellite, (2) set the appropriate radio frequency and configure other parameters based on selected satellite, and (3) provide a user-friendly interface for satellite selection and real-time control. The project targets Low Earth Orbit (LEO) satellites, such as weather satellites (e.g. NOAA, MetOp, etc.), with the potential to receive and decode signals like weather images. The system is intended to be modular; if certain features are not needed they do not break the rest of the system (e.g. for tracking celestial objects other than satellites, where decoding is not necessary).
The main purpose of this project is not to develop another universal weather satellite decoder with extra features, but a more specific and simplified system for amateur use.
]]> ]]> In today’s era, using internet platforms to convey information to others, whether family, friends, or strangers has become the norm. One of the leading social platforms in that regard is “Twitter” (now “X”). The effectiveness of communication on such platforms can be analyzed through the process of sentiment analysis. Sentiment analysis is considered a classification problem that determines whether an input is positive or negative.

The research aimed to show to what extent certain machine learning models outperform others for the given subset of data, depending on the choice of preprocessing steps within the sentiment analysis domain. This can be divided into two goals. The first goal was to present results on how one pipeline of preprocessing steps affects each machine learning model compared to the other preprocessing pipeline. The second goal was to present results on the viability of using several machine learning models for sentiment analysis of tweets by comparing the accuracies of each. For that purpose, a single subset was taken from the data and duplicated two times. Each subset duplicate had different preprocessing steps applied to it. Afterward, both subsets were fed to several machine-learning models in order to gauge their performance. 

Finally, this paper presented results on the aforementioned processes for which it was found that the Naïve Bayes machine learning model had the best accuracy, while the choice of preprocessing steps proved to be almost negligible in improving the overall model accuracy.]]> ]]>
The main intention of "TripTidy: AI-Based Travel Planner" is to make personalized trip planning simpler, less time-consuming, and more insightful for every user.

TripTidy is a web-based application designed to empower users with an intuitive and intelligent platform for comprehensive travel organization. This system enables individuals to efficiently generate personalized itineraries based on their destination, dates, budget, and preferences. Key functionalities include robust user authentication, dynamic integration with external APIs for real-time hotel (Booking.com via RapidAPI) and flight (Amadeus API) searches, AI-powered content generation (via TogetherAI), image retrieval, and geolocation services (Foursquare API) [1][2][3][4]. Users can also customize generated itineraries, track and manage expenses, and browse pre-made itineraries on a dedicated Guide page. While designed for ease of use and intuitive navigation, the application incorporates sophisticated AI logic and data processing beneath its user-friendly interface, built with React for the front-end, Node.js with Express.js for the back-end, and MySQL as the database [5][6][7][8].
]]> ]]>
The main intention of "My Wallet - AI-Based Personal Finance Manager" is to make personal finance management simpler, less time-consuming, and more insightful for every user.
"My Wallet" is a web-based system designed to empower users with an intuitive and intelligent platform for financial control. This system enables individuals to efficiently manage their transactions, track income and expenses, set and monitor saving goals, and maintain a clear overview of their financial health. Key functionalities include secure user authentication, comprehensive transaction management (both manual entry and automated parsing from uploaded bank statements), and dedicated sections for incomes, expenses, upcoming bills, and saving goals. A distinctive feature of this application is its integration of Artificial Intelligence, which intelligently categorizes transactions, detects beneficiaries from bank statement data, and generates personalized saving plans based on user-defined objectives. While designed for ease of use and intuitive navigation, the application
incorporates relatively complex AI algorithms and data processing logic beneath its user-friendly interface.
]]> The MERN stack, which features MongoDB, Express, React and Node.js, together with Tailwind is used for development. The system features secure user authentication, featureful event management and filtering, event reviewing and rating, and intuitive user management.
Eventik eases the process of event searching, enhances event visibility, simplifies participation in events and supports a more connected community, something much needed today.
]]> ]]> The aim of this project was to introduce a new solution to these issues: Lightfin, a fiscalization, loan and subsidy management system built with a microservice architecture, designed to offer enhanced performance, low delays, easy scalability and high stability.
]]>