<![CDATA[Early financial literacy education, particularly the Cinta, Bangga, Paham (CBP) Rupiah campaign, faces challenges due to the scarcity of interactive educational media capable of dynamically and adaptively assessing student comprehension. This study aims to develop a digital educational game, “RUPI”, which integrates Natural Language Processing (NLP)-based artificial intelligence to automatically evaluate students' open-ended text responses. The methodology employs a hybrid ADDIE-Scrum framework, harmonizing the rigor of instructional design standards with the flexibility of agile software engineering. The primary novelty of this research lies in its dual-evaluation architecture: the fusion of the Sentence-BERT (SBERT) algorithm and Fuzzy Matching embedded within a Non-Player Character (NPC) to process essay texts utilizing Cosine Similarity computations. This integration model empowers the NPC to deliver adaptive feedback precisely tailored to the cognitive reasoning levels and spelling tolerances of primary school children. Furthermore, the learning ecosystem is encapsulated within a robust narrative framework commencing with the introduction of an economic conflict metaphor and culminating in a title-awarding resolution to sustain students' intrinsic motivation. The results demonstrate that the integration between the frontend engine (Construct 3) and the AI server (FastAPI) operates seamlessly via asynchronous communication protocols (AJAX). Pedagogically, the interface design, coupled with visual puzzle mechanics as cognitive icebreakers, is proven to effectively mitigate cognitive load, thereby establishing an immersive, measurable, and scientifically grounded instrument for monetary literacy socialization.]]>
