M2 Master Agentic AI in virtual worlds An Interactive AI Experience for Sustainable Nutrition "A Better Bite"

Gamified Collaboration & Delegation
The prototype offers an interactive alternative to traditional educational apps by placing users in a 3D kitchen. Users can either engage in hands-on cooking by following dynamic recipes or delegate complex tasks to the AI virtual chef, creating a cooperative learning environment.

3D Spatial Awareness
The AI assistant is not just a disembodied voice - it is a spatially aware entity within the Unity environment. The virtual chef understands the layout of the kitchen, knows the locations of ingredients and tools, and can actively guide users through the physical space.

End-to-End Voice Pipeline
To enhance player immersion, the platform utilizes a complete audio processing pipeline. By integrating local Speech-to-Text (STT) and Text-to-Speech (TTS) models, users can communicate with the AI chef naturally via hands-free voice commands, receiving immediate, voiced responses.

Contextual Memory Management
The AI does not treat each input in isolation. Through advanced session management, the assistant retains a persistent conversation history. It remembers the current recipe state, previous user choices, and past questions, ensuring a continuous and coherent collaborative experience.

Multi-Agent Dynamics
The framework supports autonomous interactions beyond the player. The system features AI-to-AI communication, allowing virtual customers and the AI assistant to converse dynamically with one another, creating a lively and reactive virtual world.

Scientific & Culinary RAG Database
To support fact-based, scientifically accurate dialogue, the AI is grounded by a local vector database. This Retrieval-Augmented Generation (RAG) pipeline is populated with both practical cooking recipes and scientific research from the food4future initiative, minimizing AI hallucination and enabling detailed science communication.

Live Web Search Fallback
While the local RAG database provides verified scientific context, the system remains highly flexible. If a user asks a question outside the pre-loaded knowledge base, the AI assistant can autonomously trigger a live web search to retrieve and integrate up-to-date information into its response.