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

We are designing a low-cost support system for students in the Global South, with an initial focus on rural regions in Mexico. The goal is to help identify when students may be experiencing stress and to provide forms of support that can help them better manage it.

-To ensure accessibility in low-resource settings, we plan to build the system using FPGA-based hardware, enabling efficient, low-power, and offline-capable AI inference. This hardware layer will be complemented by integration with current generative AI APIs when connectivity is available, allowing the assistant to provide richer, context-aware support.

Rural students in Mexico, and across the Global South more broadly, often face high levels of stress shaped by poverty, long travel times to school, limited educational resources, family responsibilities, and uncertainty about future opportunities. Their stress is not only academic. It is often tied to everyday material and social pressures.

-Perspectives and Team Deliberation

+At the same time, they usually do not have the same support systems available to students in cities. Many rural communities have limited access to counselors, psychologists, extracurricular programs, reliable internet, and digital mental health tools. This means students may experience significant stress but have few resources to help them manage it.

-Within our team, we discussed several tensions:

+Their stress also differs from that of students in the Global North and urban settings. In those contexts, stress is often framed around grades, competition, and social pressure. For rural students in Mexico and the Global South, stress is more often shaped by structural inequality, economic hardship, weak infrastructure, and limited institutional support.

-Cloud vs. Edge AI: Some argued for leveraging powerful cloud-based generative AI systems for richer capabilities, while others emphasized the need for offline-first solutions due to connectivity constraints.

-Performance vs. Accessibility: We debated whether to prioritize high-performance models or systems that are affordable, energy-efficient, and deployable at scale.

-Automation vs. Augmentation: We considered whether the AI should act as a primary tutor or as a supportive companion that augments students’ existing learning processes.

+This use case matters because it focuses on students who face serious stress but are often overlooked by existing support systems. It calls for tools designed around their realities, not around the assumptions of urban or high-income contexts.

Our motivation is to design AI systems that meaningfully support students who are often overlooked by existing technologies, particularly rural students in Mexico and across the Global South. Much of today’s AI for education and well-being is built around assumptions of stable infrastructure, abundant resources, and access to professional support. We are motivated to challenge these assumptions by creating systems that reflect the real conditions students face, including limited connectivity, fewer institutional supports, and different forms of stress.

-Through these discussions, we converged on a hybrid approach: combining FPGA-based edge computing for local, low-cost inference with selective integration of generative AI APIs when connectivity is available. We also aligned on designing the system as a companion, not a replacement for teachers or human support systems.

+We are also driven by a commitment to human-centered and worker- and community-centered AI. This project extends that approach to students, focusing not only on learning outcomes but also on well-being, emotional support, and equitable access to resources.

-Proposed Contribution and Conditions for Implementation

+Our learning objectives are threefold. First, we aim to understand how stress is experienced and expressed by rural students, and how it can be responsibly assessed through lightweight AI interactions and survey-based methods grounded in existing research. Second, we seek to explore how AI systems can provide meaningful, context-aware support without over-reliance on constant connectivity or high-cost infrastructure. Third, we want to study how combining online AI systems with low-cost physical resource hubs can create more resilient and inclusive support models.

-We propose the design of a low-cost, FPGA-based AI assistant that functions as a daily learning companion for students in rural regions. The system will:

+Through this work, we hope to generate both practical system designs and broader insights into how AI can be used to support well-being in underserved communities in ways that are culturally grounded, accessible, and sustainable.

Situation or context

+We address the lack of accessible, context-appropriate support for managing stress among rural students in Mexico and, more broadly, across the Global South. These students often face high levels of stress driven by economic precarity, long travel distances to school, limited educational resources, and family or work responsibilities. At the same time, they have limited access to mental health services, school counselors, or reliable digital tools that could support their well-being.

-Operate locally on FPGA hardware to enable low-power, affordable, and offline-capable AI support

-Provide core functionalities such as tutoring, question answering, and skill-building without requiring constant internet access

-Integrate with generative AI APIs when connectivity is available to enhance capabilities

-Be developed through participatory design with students and educators, ensuring cultural and contextual relevance

+Critical analysis

+Existing AI and digital well-being tools are largely designed for urban and Global North contexts. They often assume stable internet access, individual device ownership, and the availability of professional support systems. As a result, they do not translate well to rural settings, where connectivity is inconsistent and support infrastructures are minimal. Additionally, many systems treat stress as an individual and short-term issue, rather than as something shaped by structural conditions such as inequality and limited opportunity. This creates a mismatch between the design of current tools and the lived realities of rural students.

-Conditions for implementation include:

+Team perspectives and debates

+Within our team, we discussed different approaches to addressing this gap. One perspective emphasized building a fully online AI system that could scale easily and provide personalized support. Another perspective raised concerns about over-reliance on connectivity and highlighted the importance of designing for offline access and shared community resources. We also debated how to balance automated support with human involvement, particularly in sensitive areas such as mental health. Through these discussions, we converged on a hybrid approach that combines AI-based support with pathways to human assistance and offline resources. This approach reflects a shared agreement that effective systems must be both technologically scalable and grounded in local realities.

-Access to low-cost FPGA hardware and support for deployment in rural settings

-Partnerships with local schools, educators, and communities for co-design and adoption

-Intermittent or minimal connectivity to enable periodic access to cloud-based AI services

-Safeguards to ensure privacy, autonomy, and responsible use, avoiding surveillance or misuse

+Proposed contribution and conditions for implementation

+Our contribution is a hybrid support system composed of two integrated components. First, an online AI system that engages students through conversations and survey-based check-ins grounded in stress research, enabling the system to assess stress levels and provide tailored recommendations. Second, a low-cost autonomous physical system deployed in schools or community spaces that provides offline resources such as books, games, and creative materials that support stress management.

-Under these conditions, this system can contribute to reducing educational inequities by making AI accessible in contexts where it is currently out of reach, while maintaining a strong focus on human-centered and context-aware design.