neuroguIA: A Hybrid Conversational AI Emotional Support System for Neurodiverse Groups

neuroguIA is a hybrid conversational AI system designed specifically for neurodiverse individuals, integrating NLP, context memory, adaptive dialogue routing, and supervised generative AI to provide social-emotional and functional support. The system is corely positioned as an auxiliary companion tool in educational and family settings, not a replacement for clinical or therapeutic care, emphasizing safety, interpretability, and human supervision.

Neurodiversity includes neurodevelopmental differences such as autism spectrum disorder, ADHD, and dyslexia, with approximately 15-20% of the global population having related characteristics. This group often faces challenges like sensory overload, emotional breakdowns, executive function disorders, and social communication difficulties. Traditional support relies on human intervention, but resources are limited, costs are high, and 24/7 companionship is difficult to provide; general AI assistants lack deep understanding of their specific needs and safety mechanisms, so the neuroguIA project was born.

The core design philosophy of neuroguIA is to serve as an auxiliary companion tool in educational and family settings, not a replacement for clinical or therapeutic care. The system adopts a multi-layered intelligent architecture, emphasizing safety, interpretability, and human supervision, aiming to provide social-emotional support and functional assistance for neurodiverse groups.

neuroguIA adopts a three-layer hybrid architecture: 1. Interpretable baseline models (TF-IDF, logistic regression) provide transparent support; 2. Semantic understanding layer (sentence-transformers' all-MiniLM-L6-v2 model) captures the deep meaning of conversations; 3. Supervised generative AI (optional integration with OpenAI LLM, managed through strictly controlled modules). Core functions include intent classification and function category detection, emotion and functional state recognition (e.g., emotional breakdown, shutdown state, etc.), and adaptive dialogue routing (dynamically adjusting strategies).

The memory system of neuroguIA includes: session-level context memory (maintaining current dialogue coherence), supervised memory persistence (storing important interactions after supervision), dialogue curation (reviewing historical dialogues to extract patterns), reusable adaptive responses (effective templates based on historical data), and structured interaction history (supporting subsequent analysis and optimization).

neuroguIA attaches great importance to ethical safety: it clearly states that it does not replace professional services such as medical/psychological care; data is anonymized (for datasets and records used in academic research); it uses a supervised architecture (key decisions are not delegated to LLMs); and it has built-in sound fallback logic (safe degradation when uncertain).

neuroguIA represents a direction of specialized AI for specific vulnerable groups, emphasizing domain depth (understanding the specific needs of neurodiverse groups), safety first (architecture design ensures safety), interpretability (transparent decision logic), and human-AI collaboration (auxiliary tool rather than a replacement). The project demonstrates the potential of AI in the field of social welfare, and we look forward to more similar systems serving the diverse needs of humans, providing references for cross-disciplinary fields such as AI ethics and accessible technology.