Design and development of a personalized conversational interface for a "Fantasy football app"
Loading...
Date
2026-10-12
Authors
Τουβλέλιος, Ιγνάτιος Παναγιώτης
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Fantasy Premier League (FPL) is one of the world’s most popular online fantasy sports games, in which millions of managers worldwide are required to make a series of critical decisions regarding their virtual teams every gameweek, in order to accumulate as many points as possible from the real-world performances of the footballers they select. However, the official platform primarily provides raw statistical information and limited decision-support capabilities, while existing third-party applications are mainly targeted at experienced users and frequently organise information through dense tables and long lists of metrics. Managers are therefore forced to switch between multiple websites and tools to analyse players, compare alternatives, and manage their squads, frequently through subscription-based services. They are left with a fragmented and demanding environment that imposes a considerable cognitive load and requires substantial time to learn, and that becomes difficult to use for those without much experience.
This thesis presents the design, development, and evaluation of a complete full-stack Fantasy Football web application with a personalized conversational interface (AI chatbot). The central contribution is twofold: first, it implements the game itself and comprehensively redesigns the overall user experience (UI/UX) compared with both the official Fantasy Premier League application and existing third-party platforms, bringing together team management, data analysis, player-performance predictions, and decisionsupport functionalities within a single coherent and user-friendly environment. Second, it integrates a conversational Artificial Intelligence coaching assistant that personalises its recommendations according to each manager’s preferences, playing style, and current squad status, providing tailored explanations and decision support through natural language interaction.
Underneath runs a machine-learning pipeline that forecasts points, minutes played and likelihood of appearance for the next gameweek, using gradient-boosted tree models with rolling-origin cross-validation and an acceptance check that retains the previous model whenever a retrained one performs worse. The resulting predictions are utilised by a hybrid recommendation engine that combines machine-learning techniques with heuristic rules, as well as by an integer linear programming optimiser for constructing the initial squad. A Large Language Model layer then turns these structured numerical recommendations into natural, explanatory dialogue. The platform is supported by a complete backend, a relational database, an automated data-synchronisation mechanism, a job-scheduling system, and user authentication. The implementation employs the Next.js and Express frameworks with TypeScript, Python for the machinelearning pipeline, and PostgreSQL for data management. The system was evaluated with respect to the accuracy of its predictive models, the quality of its recommendations, and the usability, clarity, and overall user experience delivered by the application.
Description
Keywords
Fantasy football, Personalized interface, User experience, Conversational interface, Machine learning, Full-stack development, Postgresql, Next.js, Express, Large language models