A growing number of individuals worldwide are diagnosed with autism spectrum disorder (ASD), with recent studies indicating a prevalence of one in 44 children. A primary challenge for many with ASD is social cognition, often described as an inability to interpret others' emotions or navigate complex social situations effectively. While traditional therapy involving one-on-one or small group sessions offers support, it is frequently limited by availability and high costs. Addressing this gap, researchers at the Institute of Human-Centered Computing at Graz University of Technology (TU Graz) have developed Simville, an affordable, accessible virtual therapy tool that complements conventional treatment methods. Published in the journal Scientific Reports, the study highlights how Simville utilizes virtual reality, advanced speech recognition, and large language models (LLMs) to create immersive, location-independent training environments. The platform allows users to practice realistic everyday scenarios, such as ordering food in a restaurant or interacting with colleagues in an office, within a safe, controlled setting. This approach enables individuals to rehearse social interactions without the fear of real-world consequences, building the confidence necessary to handle similar situations in daily life. Christian Poglitsch, who led the project as part of his doctoral thesis, emphasizes that Simville is designed to augment, not replace, existing therapeutic practices. The system incorporates gamification elements, including storytelling and immediate feedback, to encourage regular practice. A key feature of the technology is its adaptability; the number of sensory stimuli can be adjusted to match the user's comfort level, allowing beginners to start with simple interactions and gradually increase the complexity while ensuring the experience remains manageable. A significant innovation in Simville is the integration of Google's Gemini 12B large language model. This technology facilitates natural, voice-based conversations with virtual avatars. The system converts the user's speech into text, analyzes the context, and generates a tailored verbal response. The model demonstrates the ability to convey appropriate emotional undertones, creating a more authentic interaction that helps users recognize and respond to subtle social cues. Preliminary findings from a study involving 25 participants indicate promising results. After just a few training sessions, many users reported increased confidence in navigating social situations. Recognizing the potential of this technology, Simville is now being integrated into the international ETAP project led by Furtwangen University. This collaboration aims to combine the simulation interface with advanced sensor technology to dynamically adjust the intensity of the experience based on real-time user reactions. Looking ahead, the Game Lab Graz at TU Graz intends to make Simville available as a public demonstrator, offering affected individuals a practical tool for self-directed social skills training. By leveraging cutting-edge AI and virtual reality, this initiative represents a significant step toward making social skills therapy more scalable and effective for the growing ASD community.