Questanswer: um jogo de trívia com geração procedural de conteúdo baseada em inteligência artificial generativa

Abstract

Digital trivia games face a recurring longevity challenge due to the rapid exhaustion of their static question and answer sets, which compromises replay-ability and user engagement. To mitigate this limitation, this paper details the design, implementation, and validation of QuestAnswer, a mobile software ecosystem powered by Procedural Content Generation via Machine Learning (PCGML). The solution adopts a distributed and decoupled client-server architecture, consisting of a reactive mobile application developed in the Godot Engine, a .NET 9 Web API application server containerized via Docker and hosted in the cloud (Render), a relational persistence layer in PostgreSQL (Supabase), and a batch ingestion pipeline operated via a Node.js Command Line Interface (CLI) tool. This pipeline orchestrates Large Language Model (LLM) inferences through the GroqCloud API, applying prompt engineering with historical term injection to prevent duplication at the source, as well as probabilistic playful destabilization routines using ambiguous clues (trick hints). Empirical system validation was conducted with a sample of 23 Systems Analysis and Development students at IFPI Campus Picos. Results indicated high interface usability (mean 9.57/10), strong technical stability (low bug/crash incidence below 2.5/10), and broad approval of synthetic clue coherence (8.30/10) and trick hints fun factor (8.87/10). Furthermore, an A/B Test revealed that groups using larger (70+ cards) and smaller (15 cards) static card sets reported equivalent perceptions of rapid content exhaustion (7.23 vs. 7.40), empirically proving the genre’s rapid consumption speed and the essential need for a continuous supply pipeline combined with formal game design guidelines. It is concluded that the ecosystem achieved its goals, demonstrating the technical, economic, and playful feasibility of PCGML applied to mobile entertainment games.