ERRL AI Agent Demo

Description of the ERRL 0.1v Live AI Workplace Research Agent (LAWRa) The LAWR(a) is an early prototype of a live conversational AI agent developed in the Extended Realities Research Lab for use in immersive virtual reality environments. The agent is designed as an embodied professor-avatar that supports spoken interaction with users inside VR, combining real-time voice recognition, local language-model reasoning, and synthetic speech output within a VizardVR-based simulation. The core software framework of the prototype is VizardVR, which manages the immersive scene, avatar presence, interaction logic, and VR runtime environment. Within this framework, the AI agent functions as a situated academic guide: the user speaks naturally in VR, the system interprets the speech, generates a scholarly response, and returns the answer through a spoken avatar voice. Technically, the agent follows a real-time conversational pipeline: 1. Speech-to-Text: User speech is captured in the VR environment and transcribed using Deepgram’s Nova 3 speech-to-text model. This provides fast and accurate recognition of spoken input, allowing the user to interact with the avatar conversationally rather than through menus or typed prompts. 2. LLM-based reasoning: The transcribed text is passed to a local Ollama model, qwen2.5:7b, which performs the main reasoning and response generation. Running the model locally supports low-latency experimentation and gives the lab greater control over the agent’s behavior, persona, and domain-specific knowledge. 3. Text-to-Speech: The generated response is converted into spoken audio using ElevenLabs’ Eleven Turbo v2.5 text-to-speech model. This gives the avatar a fluent and natural voice suitable for live VR interaction. The agent’s persona is modeled after Marko Orel, an associate professor and workplace researcher. It is designed to mimic his academic knowledge profile, especially in relation to knowledge work, workplace transformation, remote work, hybrid work, home-office environments, and the post-pandemic evolution of work. The agent does not simply provide generic answers; it speaks as a scholarly workplace expert, offering short but conceptually dense explanations grounded in themes such as boundary management, attention fragmentation, autonomy, environmental control, digital fatigue, work design, and socio-technical change. The 0.1v build is therefore both a technical prototype and a pedagogical research instrument. It allows users to engage with workplace research through embodied conversation, while also experiencing how space, sound, domestic objects, and interruptions influence attention and productivity. In this sense, the agent demonstrates one of the core lessons of post-pandemic workplace research: productivity is not only a property of the individual worker, but also of the environment, technologies, organizational expectations, and work design surrounding that worker .