A growing number of people worldwide are living with autism spectrum disorder (ASD), with recent studies indicating that one in 44 children receives a diagnosis. A core challenge for many individuals with ASD is social blindness, characterized by difficulties in recognizing emotions in others and responding appropriately to social cues. While traditional therapy typically relies on expensive, time-intensive one-on-one or small-group sessions, researchers at Graz University of Technology (TU Graz) have developed a promising alternative. Using advanced computer game technology, they have created an affordable, accessible, and effective supplement to conventional treatment that allows patients to practice social skills without fear of real-world consequences. The innovative solution, known as Simville, is a virtual reality environment designed to simulate everyday situations such as job interviews, ordering food at a café, or chatting with colleagues. By providing a controlled setting, the system enables users to interact freely and make mistakes without the anxiety that often accompanies real-life social interactions. This immersive and playful approach is intended to build confidence and competence in a safe space, ultimately preparing users for similar encounters in their daily lives. Christian Poglitsch, a researcher at the Institute of Human-Centered Computing at TU Graz who led the project as part of his doctoral thesis, emphasizes that Simville is not designed to replace traditional therapy but to complement it meaningfully. A key feature of the platform is its ability to control the level of sensory stimuli, allowing beginners to start with simple scenarios and gradually increase the complexity as they progress. The system also ensures that training remains engaging through storytelling, clear tasks, and immediate feedback after each interaction. Technologically, Simville integrates large language models (LLMs), speech recognition, and speech synthesis to create dynamic and responsive conversations. Users can speak directly to virtual avatars, and their words are converted into text by the speech recognition system. An LLM, specifically Google's Gemini 12B model, then generates context-aware responses that reflect the avatar's personality and emotional state. According to Poglitsch, the model's ability to convey appropriate emotional undertones through voice was a standout feature, making the interactions feel remarkably natural. Preliminary studies involving 25 participants have yielded positive results. After just a few training sessions, many users reported feeling significantly more confident in social situations. The findings, which highlight the potential of virtual reality to reduce social anxiety and improve emotional recognition, have been published in the journal Scientific Reports. Looking ahead, Simville is 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 the user's real-time reactions. Furthermore, the Game Lab Graz at TU Graz plans to make Simville available as a public demonstrator, allowing individuals with ASD to access the training tool for self-directed practice. This initiative represents a significant step toward democratizing access to high-quality social skills training for the autism community.