Stellar Rune Powered By DGR AI – A Deep Learning-Powered Conversational Intelligence For Immersive 2D Gaming

Abstract

This paper presents Stellar Rune, a 2D conversational intelligence game designed to leverage deep learning techniques for an improved immersive player experience. At the core of Stellar Rune is the DGR-AI engine: a novel deep learning-based intent classification system that integrates both neural network architectures and semantic embeddings for robust natural language understanding. The proposed model leverages Sentence-BERT embeddings, which transforms the user queries into dense vector representations that preserve semantic meaning. These embeddings are then processed through a stacked architecture comprising Conv1D, Bidirectional LSTM (BiLSTM), and Dense layers, enabling the system to capture both local n-gram features and long-range contextual dependencies. A tokenization and embedding pipeline ensure efficient input pre-processing, while the neural classifier refines predictions across multiple intent classes. To further enhance reliability, the proposed system incorporates an Intent Override mechanism and a Fallback model. The override ensures correction of ambiguous or misclassified intents, while the fallback addresses out-of-scope queries using alternative matching strategies. The Stellar Rune framework is evaluated based on cross-entropy loss, intent accuracy, predicted accuracy, and actual accuracy metrics, combining correct neural predictions, semantic matches, and fallback resolutions. The proposed architecture demonstrates improved performance by unifying neural predictions with semantic similarity checks, offering a scalable solution for intent detection, natural language processing (NLP), and conversational AI. This hybrid framework, showcased through the Stellar Rune platform, highlights the effectiveness of combining transformer-based embeddings with deep learning classifiers for real-world dialog systems.