RuralWell AI: A Hybrid AI and Physical System for Supporting Well-Being in Rural Students

Team name: UNAM (Universidad Nacional Autonoma de Mexico)

Use of AI tools : We used multiple generative AI tools throughout the project. Most importantly, we integrated Generative AI API services into our intelligent interface to support interactions with the end user. Note that the AI never categorized people's stress-levels. We used instead a validated psychological survey for cateogirzing end-users depending on their stress level, particularly, we use the SISCO Academic Stress Inventory (Inventario SISCO de Estrés Académico). We also used generative AI tools during the design process. These tools helped us create early video prototypes and initial drafts of culturally aware icons and interface elements for the system. Our design team then refined these AI-generated assets using feedback gathered from people in rural communities in Mexico, ensuring the final interface better reflected local cultural contexts and user needs. We also used generative AI tools to help us to improve the grammar of. our proposal. Overall, generative AI tools supported both the technical implementation of the intelligent support system and the rapid prototyping of culturally grounded interface designs.

External feedback & contributions :
Norma Elva Chavez, UNAM Professor, mentor
Saiph Savage, Expert, Feedback, Keynote
Chinmay Das, comment and feedback
Marion G, comment and feedback
Johanna F, comment and feedback
Nicolas Lepotier, comment and feedback
Hugo Barthelmebs, comment and feedback
Caroline Beslin, Keynote
Dr. Alain Goudey, Keynote
Bastien Guerry, Keynote

Initial contribution: RuralWell AI: A Hybrid AI and Physical System for Supporting Well-Being in Rural Students

Final contribution:
Situation or context we examined:
We focused on the lack of accessible and culturally grounded mental health and educational support systems for rural students in Mexico and the broader Global South. Many rural students experience stress shaped by poverty, long travel times to school, limited educational resources, family responsibilities, and weak institutional support. At the same time, they often have limited access to psychologists, counselors, reliable internet, or digital mental health technologies. Existing AI well-being systems are typically designed for urban and Global North contexts that assume stable infrastructure and constant connectivity, creating a digital divide that excludes many rural communities.

Critical analysis:
We analyzed how current AI well-being technologies often fail to address the realities of underserved rural students. Many systems rely on unrestricted AI inference, constant connectivity, and centralized infrastructures that may not be culturally appropriate, transparent, or privacy preserving. We also identified risks related to black-box decision making in mental health contexts, where students may not understand why they are categorized in a certain way. Additionally, we recognized that overreliance on online-only systems could exclude students with intermittent internet access, further reproducing inequality between urban and rural regions.

Positions debated and arbitrated:
Within our team, we debated whether the system should rely primarily on AI-driven conversational analysis or on validated psychological instruments. We decided that the AI itself should not autonomously determine stress levels. Instead, we grounded the categorization process in the SISCO Academic Stress Inventory, a validated psychological instrument developed for Latin American educational contexts. We also debated the balance between online and offline support. While some approaches focused on fully digital systems, we concluded that a hybrid model combining AI support with physical educational resources would better address connectivity limitations and improve accessibility for rural communities. We additionally discussed concerns around privacy, explainability, and dependence on external technological infrastructures, which motivated us to explore locally deployable and open-source AI approaches.

Contributions we are proposing:
During Phase 2, our contribution evolved from an initial idea into a functional intelligent interface for supporting rural students in Mexico. The system uses the SISCO Academic Stress Inventory to guide students through a survey that categorizes their stress levels in culturally grounded ways. After completing the survey, students can interact with an AI agent that provides bounded emotional and educational support aligned with the psychological framework and stress categorization. Technically, the system connects in the backend with generative AI APIs to process student interactions and generate appropriate responses through structured prompts, predefined response rules, and stress-level-specific boundaries. The AI agent can provide reflection prompts, breathing exercises, journaling activities, study planning strategies, and emotional well-being recommendations while avoiding unsupported clinical advice.

To help bridge the digital divide, the system also includes a low-cost offline component. Based on the student’s stress categorization, the system generates a physical card that students can bring to a kiosk located in their school or community space. The kiosk then provides physical educational and well-being materials aligned with their stress category, including books, worksheets, reflection cards, mindfulness activities, games, and stress-management resources. Our Phase 2 video demonstrates this functional AI-enhanced prototype.

Overall, in this phase, we began building an end-to-end hybrid system that combines AI-based stress assessment, bounded online support, and offline educational resources in order to create a more accessible, culturally grounded, and sustainable support model for rural students in the Global South.

Explain how your contribution evolved during Phase 2:

• What feedback (comments, discussions, expert inputs, keynotes) influenced your work?

  Based on the feedback we received, we started redesigning our prototype to make the system more transparent, privacy-preserving, culturally grounded, and sustainable for rural communities. We refined the stress assessment process so that categorizations are based on validated psychological instruments from Mexico and Latin America, rather than unrestricted AI inference. We also redesigned the interface to include a tracking and explainability component where students can understand how their stress categorization was generated and how it changes over time. In response to concerns about privacy and technological dependency, we began exploring the integration of locally deployable Small Language Models (SLMs) and open-source AI tools that could run on school-managed infrastructure, reducing reliance on external servers and improving data sovereignty. Additionally, we further strengthened the hybrid online-offline design of the system so students can receive both AI-supported interactions at school and physical educational materials through the kiosk system, ensuring support even in low-connectivity environments.
  
  The keynote sessions also influenced how we refined and redesigned our system during Phase 2. The session on Project Success by Caroline Beslin helped us think more carefully about how knowledge and emotional support could be effectively transmitted to rural students through both AI interactions and physical educational materials. It encouraged us to better structure the layered support model between the AI system, psychologists, and offline kiosk resources. Similarly, Dr. Alain Goudey’s keynote on how people actually use AI in practice influenced our decision to avoid fully autonomous AI decision-making. Instead, we grounded the system around validated psychological instruments and bounded AI interactions, recognizing that understanding how humans engage with AI is as important as the AI capabilities themselves. This reinforced our focus on explainability, transparency, and student agency. The Human-Centered AI session by Dr. Saiph Savage shaped our emphasis on culturally grounded AI design. It reinforced the importance of designing systems around the realities, values, and infrastructure constraints of rural communities in Mexico, rather than adapting assumptions from Global North educational settings. This influenced our integration of culturally relevant stress frameworks, explainability mechanisms, and hybrid online-offline support. Finally, Dr. Bastien Guerry’s discussion on AI coding assistants and free software motivated us to further explore open-source AI tools and locally deployable Small Language Models (SLMs). This influenced our redesign toward privacy-preserving and locally maintainable infrastructures that could reduce dependence on external technology providers and better support technological sovereignty in rural communities.

• We want to thank everyone that helped to power our final system. Thank you to all the mentors, all the teams participating and giving feedback, all the keynotes, and all the organizers of the AI grand challenge. It has been a wonderful experience for learning and growing together :)

Video pitch: