From Assumption to Practice: Learnings from designing an equitable teacher centered- AI for low- income classrooms in India

• In Event: Teacher education and educator support in South Asia

Proposal

Introduction
In contexts marked by resource scarcity, oversized classrooms, and systemic marginalization, education often perpetuates rather than dismantles the structural violence that characterizes our divided world (Quintero and Fida, 2023). Teachers are often expected to manage sixty or more students without adequate materials or support, while students face structural barriers that make meaningful learning difficult (NCPR, 2024). These challenges not only limit educational outcomes but also reinforce cycles of exclusion (UNESCO, 2021).

In response, educational technology is often marketed as a solution for democratizing access and easing teacher burdens (Haleem et al., 2022). AI- powered tools, particularly, are positioned as interventions capable of personalizing instruction and reducing cognitive overload. Yet, evidence suggests that these promises rarely translate into practice (Holmes, 2023). Studies reveal that 80-90% of EdTech interventions in low and middle income countries fail to produce measurable learning gains, with some even correlating with negative outcomes (Mckee, 2015). These shortcomings stem less from the absence of innovation than from fundamental design misalignments. Literature points to a persistent pattern of technological colonialism (Adam, 2018), where interventions treat teachers as passive implementers and overlook local pedagogical contexts (Lurvink & Pitchford, 2023). Monolithic, one-size-fits all designs often assume standardized infrastructures and prioritize Western norms (Eppard et al., 2021) , sidelining the educators’expertise who navigate challenging conditions daily. As a result, the very tools intended to democratize learning often end up reinforcing inequalities (UNESCO, 2023). It is against this backdrop that we piloted TeachAIde, a hypercontextualised AI-powered teacher assistant that uses classroom information to provide tailored support in lesson planning, classroom strategies and differentiated instructions to help reduce the cognitive load of teachers, in three low-income schools in Delhi. Drawing on the learning that emerged through the implementation of the pilot, this design ethnography study asks: How does iterative, design-based processes of observation inform the development of more context-sensitive and equitable educational technologies?

Methodology and Analysis
This study employed a Design-Based Research (DBR) approach (Sandoval & Bell, 2004) informed by design ethnography (Müller, 2021) to examine how design assumptions were tested and reconfigured during implementation. Design ethnography enabled close engagement with teachers and students as they encountered and adapted TeachAIde, treating implementation as an ongoing negotiation between intended design and contextual constraints.
Data were drawn from field notes, implementer interviews, and implementation records across three sites: one unrecognised low-income school in Sangam Vihar, Delhi, serving migrant families excluded from formal education, and two overcrowded, resource-constrained Delhi Government schools. Field notes captured classroom observations, contextual dynamics, and researcher reflections. In-depth interviews explored local adaptations and participatory design experiences, while regular debrief meetings surfaced emerging insights and design challenges. Data were analysed using Charmaz's constructivist grounded theory (2006) through iterative coding and memo writing, allowing themes to emerge from participant experiences.

Findings
Findings and discussion are explicitly combined to show how intervention design shapes implementation and informs equity in a design-based research approach.
Finding 1: Perceived vs. Actual Reality: Although the survey was designed in Hindi and contextualised drawing on prior studies, classroom use revealed that the language remained too formal and abstract for many students. As field notes reported, "Most of the students still could not understand what some of the questions were and we had to explain it in detail." While schools reported their students as English/Hindi proficient, this typically meant the ability to reproduce memorised phrases without functional fluency. Within a DBR approach, it underscores the need for continual iteration: contextualisation is necessary, but only repeated testing and adaptation can ensure that tools align with lived classroom realities.
Finding 2: The Need for Triangulation: Before implementation, principals and teachers expressed a strong preference for English-medium materials, even where functional proficiency was limited. As one pre-intervention meeting recorded, "Even if the children cannot use it well, we still want them to learn in English." This preference continued after implementation, reflecting the symbolic value of English. As a facilitator reflected, "The demand for English is more about status than about what works in the classroom." This shows why triangulation is essential. Relying only on teacher and principal interviews risks designs being shaped by aspiration rather than classroom realities.
Finding 3: Hidden Assumptions of Autonomy: Both student and teacher engagement revealed hidden assumptions built into the design of interventions. Surveys designed for self-administration assumed literacy and independence that classrooms could not provide. As one facilitator noted, "What we thought would take 20 minutes stretched to over an hour because every child needed explanations." Similarly, TeachAIde assumed that teachers would be confident, active partners, but engagement declined without clear communication or follow-up. Field notes captured this dynamic: "The tool was demonstrated, but the teachers stepped back and let others handle it." Both students and teachers require more structured support than the tools assumed. This underscores the importance of designing not only the tools themselves but also the scaffolds around them, including ongoing support that builds confidence and agency for adoption.

Conclusion
This study began in contexts of scarcity, overcrowding, and systemic exclusion, where education often reproduces structural violence. In such settings, technology, and AI in particular, is increasingly promoted as a way to bridge social distance, strengthen equity, and foster cohesion (Bi, 2025). The findings reveal that even participatory, teacher-centered AI tools confront barriers such as linguistic inaccessibility, data poverty, and entrenched hierarchies. These challenges show that adoption cannot be reduced to tool availability but requires grappling with the inequities shaping classrooms and teacher practice. They also highlight the importance of process (DBR): assumptions that seemed sound in design often broke down in practice, and were addressed during debriefs and feedback sessions. Iteration and adaptation were not optional but essential for aligning interventions with lived realities. For technology to fulfill its promise as a bridge, it must be built through culturally responsive, context-sensitive design that elevates teacher agency and enables authentic co-creation. This research highlights that bridging equity and cohesion requires not only new tools but also design processes that confront structural barriers and adapt through continuous cycles of reflection and re-planning.

Authors

• Shaivi Soni, Aawaaz Foundation

• Vatsala Singh, Aawaaz Foundation

• Smriti R. Sharma

• Tarang Tripathi, University of California, San Diego

• Chandraditya Raj, Aawaaz Foundation