The rapid advancement of Artificial Intelligence (AI) promises significant socio-economic gains, yet over 2 billion people in “digital silence zones” remain excluded due to infrastructure gaps. Current AI ecosystems rely on energy-intensive centralized clouds or edge models requiring periodic synchronization, both of which consume massive water and energy resources for cooling and operation, failing in crisis zones where such resources are absent. This paper introduces El Faro, a self-sustaining “advanced frugal AI stack” designed for permanent-offline operation through the strategic repurposing of e-waste and local materials. The framework integrates three synergistic layers: (1) an intelligence layer utilizing 4-bit quantized models on legacy silicon; (2) a resilient hybrid mesh communication backbone; and (3) a biomimetic, clay-based passive cooling enclosure that eliminates mechanical energy overhead. Preliminary validation of the intelligence layer demonstrates that local LLM inference is feasible on constrained, recycled hardware. El Faro is presented as a speculative architectural design with early feasibility indicators; comprehensive mesh networking and thermal validation remain as immediate future work. The framework provides an architectural blueprint for inclusive innovation, suggesting that technological sovereignty and digital agency may be maintained through ecological integration, even in total infrastructure collapse.

Stress remains a significant health concern, yet commercial monitoring devices are often too expensive and complex for populations in developing countries. This research presents Spondon, a low-cost, multi-sensor wearable designed to detect stress in resource-constrained environments. Utilizing an ESP32 microcontroller with MAX30102, GSR, and temperature sensors, the device collects physiological data. We evaluated the system with 15 participants using the Stroop test to induce stress across baseline, task, and recovery phases. Statistical analysis confirmed a distinct 7.95% drop in HRV (RMSSD) during stress. The system achieved a 75.0% accuracy in binary acute stress detection (Baseline vs. Stress) using a Random Forest classifier with personalized baseline normalization. In addition, when evaluating continuous real-world monitoring across three physiological states (Baseline, Stress, and Recovery), Spondon achieved a 58.4% multi-class accuracy, successfully outperforming the 33.3% random-chance baseline. This work contributes a validated, accessible hardware platform and methodology, democratizing stress research for low-income communities and highlighting the importance of personalized, context-aware design in HCI.

When a public agency routes citizen queries through a commercial language model, the accountability questions that go with that — whose data is it, how was the output produced, what happens when a citizen gets a wrong answer tend not to get answered. This poster argues that those are not edge cases. They are routine requirements of public-service delivery, and the prevailing large-model default is structurally unprepared to meet them.

I propose the Citizen Service SLM Architecture (CSSA). Rather than bolting governance onto an otherwise capable model, CSSA treats sovereignty, auditability, and escalation as design concerns from the start: the agency controls the model and its knowledge base, responses trace back to sources it curates, and the interaction inherits the same review path as any other official act. The architecture separates citizen-interface, AI-processing, and backend-integration concerns, with grounding and oversight mechanisms running across each layer.

The design is derived from a structured literature synthesis across digital government, retrieval-augmented generation, and algorithmic accountability, and evaluated against four government service scenarios. Multilingual query resolution is the primary case: it shows how the architecture can extend access to linguistically underserved citizens without surrendering institutional control over policy interpretation. For many citizen-facing settings, a smaller model grounded in what the agency already knows and governed inside its existing workflows fits public-service obligations more closely than a general-purpose model hosted elsewhere.

Early childhood education in Ghana emphasises literacy and numeracy at the expense of social-emotional learning (SEL), leaving many preschoolers under-prepared to navigate cooperation, empathy, and conflict. Existing SEL technologies are designed for Western contexts and rarely reflect Ghanaian cultural realities, while structured SEL approaches such as Montessori remain confined to fee-paying urban schools. This paper presents Sode, a progressive web application that scaffolds SEL for children aged four to six through 3D cooperative scenarios, parent-child empathy mini-games, and a progress dashboard, all anchored in Ghana’s Early Learning and Development Standards. We describe the design rationale, the system architecture, and a formative evaluation conducted with children from traditional and STEM-oriented schools in Accra. Thematic analysis of the evaluation surfaces four design implications for culturally grounded SEL technology in low- and middle-income contexts. The contribution to COMPASS is a worked example of how computing can address an under-served dimension of early childhood education in the Global South.

Museums are sites of cultural participation where participation depends on interpretive agency. Audio description is the primary mechanism through which blind and low-vision (BLV) visitors access visual art, yet it is rarely examined as a system of power. This paper argues that dominant audio description practice, as reflected in our corpus, does something more consequential than provide access: it resolves interpretive ambiguity in advance, positioning BLV visitors as recipients of pre-authored meaning rather than as agents of their own cultural encounter. Through a qualitative analysis of fourteen audio descriptions from major institutions, including Tate Modern, MoMA, and the Guggenheim, we identify three structural mechanisms through which interpretation is fixed before the visitor arrives: resolution as ordering, certainty, and omission. Within this corpus, 80% of descriptions include emotional or interpretive language, and only 20% preserve ambiguity. These patterns that suggest structural tendencies in mainstream institutional practice rather than prevalence estimates across the field. We introduce negotiable interpretation as a design paradigm that redistributes interpretive control, and describe a conversational prototype through which BLV visitors can shape how meaning is constructed during the encounter. We argue that accessibility in cultural institutions is not merely a technical problem of information transfer, but an equity problem of who holds authority over meaning; and that technology can either entrench or challenge that asymmetry.

QR-based digital payments are increasingly used by small vendors in Nepal, but confirming whether a payment has actually arrived remains difficult during busy shop interactions, weak connectivity, and customer turnover. This paper presents a formative pilot study with 10 small vendors to understand how they verify QR payments and what breakdowns they experience with current confirmation methods such as SMS alerts, app notifications, customer screenshots, and soundboxes. Our findings show that vendors rely on multiple fragile signals to confirm payments, including customer-provided proof and trust, and that delayed notifications, poor connectivity, suspicious screenshots, and high customer traffic make verification unreliable. Based on these findings, we derive design requirements for a Smart VoiceBox prototype: clear Nepali audio, hands-free confirmation, weak-connectivity tolerance, low cost, battery support, and reduced reliance on customer-provided proof. We present a proof-of-concept prototype that uses GSM/SMS-based triggering and local audio playback to announce payment confirmation aloud. The study contributes early design evidence for accessible, voice-based payment confirmation tools for small vendors in Nepal.

Rapid urbanization and inadequate waste segregation in cities such as Kathmandu have intensified pressure on landfills and reduced the recoverability of recyclable materials, affecting both municipal sustainability efforts and informal waste work. This paper presents a low-cost edge-AI prototype for waste classification and sorting in resource-constrained urban contexts. The prototype integrates a lightweight MobileNetV2 convolutional neural network with a Raspberry Pi 4 and simple actuation hardware to support cloud-independent waste classification. Using a locally collected dataset, we evaluate how offline data augmentation can address severe class imbalance in early-stage waste classification models. This strategy improved organic waste recall from 0.22 to 0.96 and increased overall accuracy to 93.60%. We also demonstrate initial hardware integration, where model predictions are used to actuate physical sorting. Rather than presenting the system as deployment-ready, we position it as a technical feasibility prototype and discuss the design considerations needed for future field evaluation with municipal stakeholders, informal waste workers, and users of public disposal sites. The work contributes an initial step toward context-aware edge-AI systems for sustainable waste management in Nepal and similar low-resource settings.

Applications with Short-Form Videos (SFV) like TikTok and Instagram have substantial popularity among university students. The constant usage of these applications sparks discussions about the potential decline in their cognitive functioning, especially task attention. The Bangladeshi context of such app use is distinct in the sense that the algorithm-driven For You Page (FYP) generates videos in multiple languages, often in rapid succession. We explore if such frequent context switching contribute to the reduction in attention and the attendant cognitive fatigue. Moreover, the conventional form of sociality for Bengali culture, Adda, seems to be undergoing a transition towards a more passive, screen-mediated sociality, shaped by doomscrolling (a continuous, inescapable scrolling habit). We present a qualitative study, based on semi-structured interviews, intended to capture the lived experiences of Bangladeshi university students constantly exposed to multilingual short videos on social media, particularly their loss of attention, attendant cognitive fatigue, and the shifting dynamics of sociality. By foregrounding students’ lived experiences, this study contributes to HCI research by showing how constant switching between multilingual short videos shapes attention, cognitive burnout, and culturally rooted social interactions in a multilingual Global South context.

The initial experience of a smartphone is a pivotal point in developing users’ first impressions of a device. During this onboarding process, users can adjust language settings, make or log into account, manage privacy settings, join networks, and customize device settings. Although smartphones are common, some users are still uncertain about setting them up, particularly during the initial configuration process when their digital devices are unfamiliar. There are differences in their level of literacy and they require assistance from friends, and mobile phone retailers. This paper analyzes smartphone configuration experiences among participants in Bangladesh through an anonymous online survey (n = 20). The survey investigated who usually performs smartphone setup, how difficult users found their last setup experience, which steps felt confusing, whether users understood permission requests, whether they skipped ambiguous steps, and what could make setup easier. This analysis reveals that most participants were able to set up without assistance but found creating accounts and authentication to be the most challenging steps. In this paper, authentication refers to sign-in or credential-related activity within Google/Apple account setup and was not measured as a separate survey item. Open-ended responses also indicated the requirement for step-by-step guidance, simplified explanations, Bangla or local language assistance, and visual instructions. The results offer some preliminary guidance in aiding the design of more straightforward and user-friendly smartphone onboarding experiences.

AI research in healthcare often promotes disruptive innovations for diagnosis and care, yet the infrastructural and clinical realities of many public health systems raise questions about whether such transformations are sustainable. We present an exploratory qualitative study with ten Parkinson’s disease (PD) specialists and decision-makers in Ecuador’s public healthcare system—a low-resource setting—examining how they perceive disruptive AI approaches for PD. Our findings show that infrastructural constraints shape not only AI adoption, but also stakeholders’ capacity to imagine disruptive technological futures. Participants favored AI systems that support existing clinical workflows over stand-alone predictive systems for early diagnosis. We discuss methodological and ethical implications for responsible, sustainable AI design and argue for revalorizing non-disruptive, support-oriented AI as a legitimate goal for low-resourced public healthcare systems.

Environmental monitoring infrastructure is not neutral. The communities bearing the highest urban heat burden, fence-line neighborhoods like Little Haiti, Miami, remain structurally absent from the data systems that could compel intervention on their behalf. This paper presents ClimateSignal Logger, an open-hardware environmental data logger co-designed with Little Haiti residents and community partners through a two-year participatory process. Five students from Little Haiti and Allapattah co-designed the device in Fall 2025, conducting community interviews, testing enclosure prototypes, and identifying deployment sites alongside community partners. Students visited in Winter 2025 to advance the hardware design. The physical device was fabricated in February 2026 and the complete data pipeline validated in March 2026. Full community deployment in Little Haiti is planned for Summer 2026. ClimateSignal Logger addresses four structural gaps in available heat monitoring tools—Wi-Fi and power dependency, researcher-mediated data access, proprietary data custody, and cost barriers to dense deployment—through an ESP32-S3 microcontroller with a BME688 multi-parameter sensor and PA1010D GPS module, logging temperature, humidity, barometric pressure, and gas resistance every 30 seconds with GPS-verified coordinates, syncing via Wi-Fi or BLE to a real-time community dashboard with one-click CSV download. The 3D printed enclosure, decorated by student co-designers, is community-designed and community-named.

Women’s participation in the economy is vital to achieving gender equality, reducing poverty, and promoting sustainable development. In Bangladesh, many women possess valuable skills such as tailoring, handicrafts, baking, and cultural artistry, but religious, traditional, or social restrictions prevent them from generating income from these skills. Barriers such as limited mobility, lack of digital literacy, and lack of market access further prevent them from engaging in entrepreneurship. This study addresses this gap by capturing user perspectives to define requirements for an inclusive platform framework. We conducted semi-structured interviews with 37 participants across four regions of Bangladesh. Key findings indicate a complex pattern of conditional support functioning as soft restriction, high demand for a hybrid learning–marketplace model, and a gap between existing tools and user expectations. Based on these insights, this paper proposes a conceptual framework that prioritizes user-defined inclusivity, providing guidance for developers to move from universal to inclusive design.

Early-stage game design ideation requires designers to infer what might be true of a fictional world for its elements to coexist — a form of abductive composition that most creativity support tools (CSTs) leave unsupported. We present MOTIVUM, a browser-based non-generative CST that operationalizes Stith Thompson’s Motif-Index of Folk-Literature (1955-1958). The tool draws three folklore motifs from Category D: Magic and invites the designer to hypothesize a world logic that would render the motif triad coherent, while retaining all creative interpretation with the human author. The study reports the concept defining and concept appropriating phases of an ongoing concept-driven interaction design (CDID) process. Grounded in an abductive framework that treats the interface as an eco-cognitive environment, the system encodes the fill-up problem as a structural constraint. The tool runs without installation or external infrastructure and can be fully localized by any language community through plain-text replacement of its data layer. An exploratory comparison of LLM-generated world-building outputs under abductive versus non-abductive motif development conditions are presented. These sessions are used to ground a subsequent concept-evaluation phase, for which empirical validation with independent designers remains future work. We suggest that folklore motifs, with their granularity and evocative qualities, may be well suited to structuring conditions for human abductive reasoning, and we present MOTIVUM as groundwork for a subsequent concept evaluation phase.

Emotions, engagement and well‑being are fundamental to primary students’ learning. Existing work on craft-based interventions and social robots shows benefits for creativity, mood and engagement. However, they rarely target young disaster‑affected learners or integrate craft and AI companions to support socio‑emotional recovery, such as community‑level trauma. This study investigates Loving Baby, an AI‑supported robotic companion embedded in a craft workshop for 96 primary students. The robot provides self-determination theory (SDT)‑informed emotional and scenario support while students design tailored clothes and write heartfelt words. SDT is administered pre‑ and post‑ workshop and shows significant improvements (p <.01) in behavioural, emotional and cognitive engagement, and motivation (p <.05). Internal consistency is excellent (Cronbach’s α =.955), with strong inter‑correlations between subscales (r =.78–.88), indicating a coherent engagement construct. Students also reveal rich emotional labelling (joy, sadness, anger, fear), links between happiness and helping others, and perspective‑taking towards peers and Loving Baby. The findings provide initial evidence that an AI‑supported robotic companion with a craft workshop can enhance engagement and motivation while opening a safe space for emotional expression in a post‑disaster primary context.

Urban air pollution is a major public health and sustainability concern, yet there is a limited understanding of how air-quality monitoring is incorporated into users’ everyday routines. Drawing on social practice theory, we examine monitoring as a configuration of materials, competences, and meanings, and report results from an exploratory survey of end users (n = 91). We identify two monitoring configurations: situational monitors, who report infrequent monitoring and lower indicator familiarity, and tech-engaged monitors, who report more regular monitoring and greater engagement with monitoring tools. Building on these patterns, we extrapolate design implications for wearable air-quality monitors (WAQMs), including support for different monitoring rhythms, design for interpretation and situated action, and bodily and domestic integration. We position this work as exploratory and call for qualitative and longitudinal research on monitoring-as-practice.

Unmanned Aerial Vehicles (UAVs) enable scalable and low-cost environmental monitoring. However, their deployment is still challenged by complex natural environmental conditions in the real world. This study employs a controlled experiment to investigate the impact of litter type, ground type, and illumination on litter detection performance in a forest park. A dataset of 375 aerial images was collected by a UAV, and a YOLO-based object detection model was trained and evaluated. ANOVA was used to test the impact of litter type, ground type, and illumination conditions on litter detection accuracy. The results show that litter type is a statistically significant factor, with significant interaction effects between litter type and ground type. Ground type and illumination alone do not exhibit significant effects on detection accuracy. Additional subgroup analysis revealed that composite environmental conditions influence detection accuracy for certain litter types. The findings indicate that detection performance is influenced by interactions between litter types and certain environmental conditions, rather than by individual factors alone. Based on the findings, three deployment-oriented strategies are proposed, including litter-type-aware monitoring, interaction-aware environmental planning, and composite condition-aware model training. These strategies can help balance automated UAV-based detection and human intervention, as well as support proper data sampling design.

We contrast assumptions embedded in AI/ML systems to patterns of repetition in discussions about weather-related risks amongst members of remote First Nations communities. Natural language processing (NLP) is increasingly proposed for use in emergency management. Thus, we analysed how repetition worked as a collaborative practice when participants in First Nations-led workshops in Australia’s Northern Territory talked about planning, responding to, and recovering from cyclones, floods and other extreme weather. Repetition supported shared understanding between languages, reinforced relational accountability, and sustained attention to unresolved issues. Yet, AI systems such as Large Language Models (LLMs) treat repetition as redundancy through dataset deduplication, alignment and preference optimisation, summarisation and compression. This can amplify extant patterns of non-listening in colonial institutions that require First Nations communities to bear the burden of ‘repetitive epistemic labour’ [29] and restate their concerns and knowledge. We conclude that sensitivity to how repetition can suppress meaning is essential in AI/ML design and engineering, especially for intercultural settings where communication is simultaneously informational, relational, and accountable.

Transport planning relies on estimates of how commuters value their time, but it is challenging to measure these preferences, especially in resource-constrained environments. Standard approaches rely on how much people say they value their time, and produce biased estimates. Revealed-preference methods based on smartphone or platform data are often unavailable or unrepresentative of underserved populations.

We present a method to measure willingness to wait by combining an SMS-based platform accessible via basic mobile phones with human enumerators who provide limited oversight. The system delivers randomized monetary offers for additional waiting time and records acceptance decisions in situ. To ensure compliance, the platform uses time-varying authentication codes. This design enables large-scale, location-verified experiments in environments where digital infrastructure is limited.

We deploy the platform in Lagos, Nigeria, across 18 bus stops, engaging 640 commuters and generating approximately 600 daily interactions (message-response pairs). The system operates reliably under real-world conditions, despite intermittent connectivity and user error. This proof of concept demonstrates that low-tech systems can enable rigorous revealed-preference measurement among populations typically excluded from digital data sources.

Chatbots are being widely used to support emotional well-being, yet many are designed from secular perspectives that overlook spiritual and religious values. This paper examines how Muslim users perceive spiritually grounded emotional support through AI. Through qualitative interviews with 17 participants, we find that users seek systems that are not only empathetic but also theologically grounded, emotionally responsive, and sensitive to privacy and credibility. Chatbots are envisioned as spiritually aligned companions embedded in everyday practices. We frame this role as a mediated religious agent, whose credibility comes from the religious authority it represents. These findings contribute to design considerations for faith-aligned mental health technologies that move beyond generic support toward contextually meaningful care.

The rapid growth of large language models (LLMs) is reshaping access to health information, yet their role in sexual and reproductive health (SRH)—a highly stigmatized and sensitive domain is a nascent area of research, particularly in the Indian context. This study investigates how individuals perceive and engage with an LLM-based chatbot for SRH, examining trust boundaries, privacy expectations, and the cultural dimensions of LLM-mediated health communication. We designed a Retrieval-Augmented Generation (RAG) chatbot for sexual health and conducted a user study in which fifteen participants interacted with the system, followed by semi-structured interviews. Our findings reveal three interconnected themes: First, LLMs serve as low-stakes knowledge companions that reduce the cognitive and social burden of SRH information-seeking; secondly, privacy needs are relational and situational rather than uniform, shaped by domestic contexts and family structures; and lastly, LLM responses exhibit systematic gender bias in SRH framing—a pattern which carries significant implications for equitable chatbot design. We contribute design recommendations for culturally sensitive, privacy-aware, and gender-equitable LLM systems for SRH communication.

Voice user interfaces (VUIs) are increasingly mainstream but remain challenging to use for people with speech and communication disabilities, accented or disfluent speech, and other forms of speech variability. While previous work has explored technical approaches to improve VUI accessibility, less is known about how prepared designers feel to design accessible voice interactions in practice. We conducted an exploratory online survey (n = 22) and focus group (n = 7) to discuss designers’ accessibility practices, confidence, and perceived support needs for accessible VUI design. Survey results indicate participants were familiar with general accessibility principles and guidelines (mainly GUI-focused), yet felt little confidence and limited support for VUIs specifically. Focus group discussions highlighted organisational constraints, limited voice-specific guidance and evaluation methods, and the need to include people with lived experience and domain experts (e.g., speech therapists) in design discussions. We flag opportunities for further HCI research on tools and processes for accessible VUI design in practice.

Online video platforms have become an important resource for learning technical skills, yet their role is often understood as supplementary to formal education. This paper examines Bangla-language developer vlogs in Bangladesh and argues that, in resource-constrained contexts, such content functions as a primary pathway into software development rather than an optional aid. Drawing on a mixed-methods study combining learner survey data, creator interviews, and vlog content analysis, we show how Bangla-language developer vlogs support technical learning where formal guidance, English-language resources, and local mentoring are limited. We further demonstrate that language operates as a structural condition of access, enabling learners to engage with technical concepts that would otherwise remain inaccessible. Yet this informal learning infrastructure depends on creator labor that remains financially, technically, and socially vulnerable. By reframing developer vlogs as infrastructure rather than content, this work contributes to ongoing discussions in HCI and ICTD about access, participation, and the design of equitable learning systems in the Global South.

We present the design, preliminary deployment, and user evaluation of a Large Language Model (LLM)-based tool to improve privacy in online communications within grassroots and frontline service-providing community organizations that interact with clients as a part of their work. During coordination and service provision, personally identifiable information (PII) may be shared between workers, volunteers, and clients. Collaborating with the Seattle Community Network (SCN) volunteer organization, we built a tool that automatically redacts PII from client coordination messages shared in their public chat server, and reveals the PII only to authorized users upon request. User feedback on this early iteration reveals a high valuing of its privacy-preserving functions, but concerns around trusting LLM output, a requirement to self-host LLMs even on limited hardware, and potential frictions around redaction time and tool onboarding.

Hospital websites are necessary infrastructures for users who want to book doctor appointments, look for essential information, schedule tests, and communicate with hospital authorities. Often, the experience with hospital websites is hindered by poor usability, accessibility issues for specialized need users, outdated user interface and technical problems. For this research, we conducted a mixed-method study with public and private hospital website users in Bangladesh to evaluate their experiences and encountered difficulties with the usability and accessibility of these websites. Our findings suggest that private hospital websites have comparatively better usability, but both type of websites do not meet accessibility requirements, pointing to the need for better design interventions. This study adds to HCI research by emphasizing user insights to improve healthcare experiences in Bangladesh.

Platform laborers play an indispensable yet hidden role in building and sustaining AI systems. Drawing on an eight-month ethnography of Bangladesh’s platform labor industry and inspired by Gray and Suri, we conceptualize Ghostcrafting AI to describe how workers materially enable AI while remaining invisible or erased from recognition. Workers pursue platform labor as a path to prestige and mobility but sustain themselves through resourceful, situated learning—renting cyber-café computers, copying gig templates, following tutorials in unfamiliar languages, and relying on peer networks. At the same time, they face exploitative wages, unreliable payments, biased algorithms, and governance structures that make their labor precarious and invisible. To cope, they develop tactical repertoires such as identity masking, bypassing platform fees, and pirated tools. These practices reveal both AI’s dependency on ghostcrafted labor and the urgent need for design, policy, and governance interventions that ensure fairness, recognition, and sustainability in platform futures.

Reinforcement Learning from Human Feedback (RLHF) has become the dominant paradigm for aligning large language models (LLMs) with human values, yet its environmental cost remains nearly invisible in both academic discourse and practitioner tooling. This paper presents the first phase-resolved empirical characterization of RLHF’s carbon footprint, combining Marginal Operating Emission Rate (MOER) based carbon accounting, real-time grid carbon intensity (CI) signals, and a multi-region 90-day deployment study, a scope and resolution not addressed by prior machine-learning (ML) carbon measurement work. We find that the Proximal Policy Optimization (PPO) phase alone accounts for 59–64% of total training emissions across all model scales, reaching 78.9 kgCO₂eq for a single LLaMA-65B alignment run. We introduce CarbonAware-RLHF, an open-source scheduling framework that adaptively pauses, reorders, or temporally shifts RLHF phases toward low-carbon windows using real-time CI forecasts. In a 90-day deployment across six United States (US) Independent System Operator (ISO) grid regions (312 jobs, three independent seeds per condition), CarbonAware-RLHF achieves a mean carbon reduction of 38.9% with only a − 0.22 pp (percentage-point) reward model accuracy degradation and no statistically significant change on MT-Bench safety evaluation, the best carbon-quality trade-off among all evaluated strategies. Carbon savings correlate strongly with grid CI variability (coefficient of determination R² = 0.91), providing a practical deployment heuristic. We discuss geographic equity, carbon accounting standards, and the Jevons paradox as structural challenges that technical scheduling alone cannot resolve. All data, code, and reproducibility artifacts are released at https://github.com/Saher-Elsayed/carbonaware-rlhf.

We present SpheriCity, an expert-grounded conversational prototype designed to support trustworthy knowledge sensemaking from sustainability reports. City-level circularity assessment reports contain rich information about materials, infrastructure, and policy interventions, yet their length and heterogeneous structure make cross-document synthesis and comparison difficult for practitioners and researchers working on circular economy initiatives. While large language models (LLM) promise faster knowledge access and synthesis, their opaque reasoning, hallucinations, and lack of source transparency introduce risks for trust and interpretability, and require verification in high-stakes sustainability contexts. SpheriCity addresses these challenges through a provenance-first conversational agent that foregrounds evidence traceability, structured synthesis, and interaction scaffolds to support exploratory querying and cross-document synthesis across sustainability reports. We conducted a formative expert review with six sustainability experts using representative queries spanning cross-city comparison, policy summarization, and recommendation-oriented tasks. Experts evaluated responses across dimensions, including relevance, contextual fit, accuracy, neutrality, and depth, and provided qualitative reflections on the system’s usefulness for sustainability knowledge work. Our results reveal that transparent sourcing, contextual explanation, interpretability, and alignment with expert workflow strongly shape expert trust and judgments of system usefulness. This work contributes (1) a conversational prototype for sustainability knowledge sensemaking, (2) an expert-grounded evaluation framework for assessing AI responses in high-stakes knowledge domains, and (3) design insights into how provenance, uncertainty communication, and integration in workflow influence expert users’ trust in AI assistance for sustainability decision support.

Mobile Financial Services (MFS) have rapidly expanded digital service access in growing regions of internet use, yet low literate users remain systematically excluded due to text heavy interfaces, complex security features, and a lack of inclusive design. Through semi-structured interviews with 40 low literate MFS users in Bangladesh, we identified critical barriers: frequent transaction errors (72.5% prevalence), agent dependency (47.5% of users relying primarily on intermediaries) and PIN sharing (32.5% prevalence), and usability challenges from conventional app interfaces. Informed by these findings, we designed and developed EKASH, a simplified MFS application featuring icon-driven navigation, multilingual audio guidance in Bengali, and mandatory re-confirmation before each transaction. A task-based usability evaluation with 43 low literate participants showed 86.5% average task completion and 89% user satisfaction. Notably, among the subset of participants who verbally reported an inability to use current MFS platforms without assistance, 34% were able to complete tasks independently using EKASH, and 76% completed their transactions entirely error-free. We discuss design implications for smartphone enabled inclusive FinTech in low-literacy, low-resource contexts.

Frame analysis of migration news is a socially consequential task: media scholars and researchers who study how migration is narrated need tools that are not only accurate, but transparent, auditable, and accessible within the resource constraints typical of academic research groups. Existing LLM-based approaches rely on proprietary APIs and large models that raise concerns about data privacy, reproducibility and equitable access among media researchers. This work studies how a locally deployable open-source LLM can support interpretable frame analysis as an assistive tool. We introduce a Structured Chain-of-Thought (SCoT) prompting approach using Llama3-8B, enabling step-by-step justifications grounded in predefined framing categories. This structured design allows users to audit model outputs and examine alternative interpretations in a task that is inherently subjective. We evaluate our approach on a dataset of migration-related news and show that SCoT improves classification performance over zero-shot and few-shot baselines while remaining feasible on a single GPU. Then, we conduct a human-centered evaluation in which annotators assess the coherence and influence of “the model’s reasoning”. Results indicate that SCoT explanations are generally perceived as logical (mean score 4.1/5, though with notable variation across texts) and can prompt reflection on initial interpretations, even when disagreement persists. Our findings highlight both the potential and risks of LLM-assisted frame analysis. While structured reasoning can increase the traceability of model outputs and support critical interpretation, it can also influence human judgment in subtle ways. By enabling local deployment and emphasizing human-in-the-loop interaction, this work contributes to discussions on responsible and accessible computational tools for the study of socially impactful media narratives.

Amidst escalating global concern about the intensification of bushfire (wildfire), we investigate how interactive, data-driven articles shape people’s engagement with this expanding environmental crisis. We build on critical discourse that positions climate informatics products as sociopolitical actors, as well as research on polarisation in environmental communication, through an empirical study of bushfire news coverage and its reception by urban Australians. We conducted semi-structured interviews with 7 fire science experts and 12 city-dwellers, who engaged with three interactive news articles. Our analysis highlights the challenges of navigating uncertainty, polarised and shifting discourse, and the emotional burden of bushfire communication. We show that engagement is shaped by a dynamic interplay between affective and cognitive sense-making, which can both motivate and inhibit attention. Participants expressed a desire for actionable, place-relevant information, yet often encountered barriers in interpreting complex or ambiguous visualisations. Based on these findings, we identify design opportunities for interactive articles that (1) respond to audiences’ existing exposure to environmental discourse, (2) scaffold engagement with uncertainty, (3) connect audiences with distant people and places, and (4) support pathways to collective action.

Digital and computing infrastructures are increasingly vulnerable to the intensifying effects of climate change, with extreme events like hurricanes, wildfires, and winter storms disrupting communication systems for extended periods. These disruptions reflect longstanding inequities in how infrastructures are built, maintained, and distributed. While sustainable computing research has begun to address environmental impacts through energy efficiency and alternative system design, less attention has been paid to how infrastructures fail under climate stress and how these failures are unevenly experienced. In this paper, we present a reflexive ethnographic case study of designing a community-based emergency communication network with a mutual aid community in New Orleans, Louisiana. We use friction as an analytic lens to examine the challenges and tensions that arise when building alternative infrastructures within existing systems. Our findings identify three forms of friction: infrastructural, spatial, and epistemic. We examine how they emerge through interactions with institutional systems, urban environments, uneven expertise, and material dependencies, and how they become generative forces for reimagining relations to infrastructure. These frictions emerge through interactions with institutional systems, urban environments, uneven expertise, and material dependencies. Rather than treating friction as a barrier to be resolved, we argue for “staying with friction” as an ethical and methodological commitment by attending to how friction shapes what can be built, how systems are sustained, and how alternative infrastructures are imagined. For sustainable computing, this means moving toward confronting the histories of disinvestment, exclusion, and infrastructural neglect that shape whose systems fail and whose labor is required to ensure connectivity. By attending to friction, we show how community-based infrastructures are not designed outside existing systems, but are negotiated within them, opening possibilities for more socially just forms of computing.

Despite the importance of the Information Technology (IT) sector for sustainability, there is a lack of guidance for including Education for Sustainable Development (ESD) in IT degree programs. This research investigates perspectives of staff, students and industry, and contextual factors, to develop strategies for embedding sustainability in IT curricula. We conducted a case study at an IT department in a major Australian university, through semi-structured interviews and workshops with academic staff (n=10), students (n=28), and industry professionals (n=5), and by analysing documentation such as university plans, handbooks and company reports. A thematic analysis was used to synthesise perspectives, and sustainability topics were mapped to IT knowledge areas. While IT professionals have increasingly important roles in addressing sustainability challenges, many IT staff and students are unaware of, or uncertain about, sustainability’s relevance to their field. Additional barriers to integrating sustainability into IT programs include lack of educator expertise, overcrowded curricula, and the lack of tailored resources and support. We provide an empirical account of stakeholder perspectives on integrating sustainability into IT curricula in Australia. We offer a framework and recommendations to address these issues through loose, program-level coordination, training and resources to empower educators, paving the way for more profound change. This case study offers valuable insights for integrating sustainability into IT curricula, along with a framework and strategies that can serve as useful starting points for universities to take practical, educator-led steps towards ESD.

This paper examines how algorithmic accountability in Canada’s visa system is articulated institutionally and experienced by applicants across borders. We analyzed Immigration, Refugees and Citizenship Canada (IRCC)’s Algorithmic Impact Assessment (AIA) for the temporary resident visa (TRV) triage system using the algorithmic decision-making adapted for the public sector (ADMAPS) framework and analyzed Reddit discussions among applicants using a mixed-methods approach. We show that while institutional artifacts emphasize transparency, procedural safeguards, and bounded impacts, applicants engage in collective sensemaking to interpret opaque decisions, often relying on peer knowledge amid uncertainty. We identify three asymmetries between how institutional accountability is structured and how people perceive the process: epistemic asymmetry in access to decision logic, jurisdictional asymmetry in exposure shaped by geopolitical positioning, and temporal–relational asymmetry in how waiting and uncertainty are experienced. We emphasize why it is important to shift attention from institutional design to the uneven distribution of experiences with public-sector algorithmic governance. Together, these contributions demonstrate how algorithmic governance systems in the context of transnational migration produce structured asymmetries not captured by institutional disclosure frameworks, and how extending ADMAPS can account for those uneven translations of accountability.

This study is based on a community-led participatory soil health workshop with 34 Caribbean temporary farmworkers in Ontario, Canada. Working with two migrant community leaders as co-designers and facilitators, we centered this farmworker community as leaders of their own digital agricultural education by leveraging artifacts like tomato plants as transnational boundary objects for dialogical learning between farmworkers’ existing agricultural knowledge, and soil science principles and digital sensing instruments. We find that: (1) community-led facilitation produced unplanned but generative knowledge translation work to center farmworker perspectives; and (2) participants engaged with soil sensing principles and tools selectively, integrating new knowledge based on their own contexts and values. Through this work, we offer four workshop design insights for community-led digital education, showing how infrastructural concepts like boundary objects and infrastructural inversion can function as pedagogical tools whose force depends on dialogical conditions that begin from participants’ existing knowledge and values. We also leverage the concept of selective appropriation to consider how participants negotiate technological adoption in agricultural contexts under infrastructural constraint. Across these contributions, our aim is to support communities to build epistemic agency, which we define as a liberatory force that enables community participants to evaluate new knowledge on their own terms, in relation to their own worlds.

Understanding bird migration across full annual cycles is critical for effective conservation. The recent BirdFlow framework allows researchers to model population-scale migration without physical tracking tags by inferring likely flight paths from weekly population snapshots (such as eBird abundance maps). However, the original framework is Markovian, meaning it assumes a bird’s next move depends only on its current location, completely ignoring its history. Because of this short-term focus, the model cannot capture essential year-long biological behaviors, most notably annual site fidelity, where birds return to the exact same breeding or wintering grounds year after year.

In this paper, we extend this framework by introducing “loop” and “nested loop” graphical models. These new structures explicitly connect distant time steps, allowing the model to remember and enforce long-range dependencies across an entire year. In general, graphical model inference and learning become exponentially more expensive as graph structures become more complex. For this type of spatial modeling, the key parameter is n, the number of grid cells in the map. To solve this, we developed custom algorithms that keep the initial model-training cost manageable (O(n³)) while ensuring that generating forecasts and simulating tracks remains feasible for interactive use (O(n²)). Empirical evaluations on tracking data from four North American bird species demonstrate that our proposed models successfully enforce annual site fidelity without compromising short-term forecasting accuracy. Ultimately, these new models provide a highly scalable, biologically realistic tool for tracking how bird populations move and interact across different seasons.

Air pollution poses significant health, social, and economic risks, necessitating effective communication to promote awareness and behavioural adaptation. This study explores how combining behavioural and science communication strategies shapes stakeholder participation in co-designing air quality (AQ) communication technologies and how stakeholders engage with, represent and negotiate environmental data and issues. Data was collected across 11 workshops involving 169 participants of different ages, combining education, storytelling, user testing and participatory design to foster learning and dialogue. Findings highlight that AQ communication is entangled with questions of data usability, trust and environmental literacy. Participants emphasised the importance of audience-specific AQ messaging, integrating AQ technical data with comprehensible visualisations, and combining persuasive communication with evidence-based information. Tensions emerged between individual adaptive behaviours and structural barriers, reinforcing the socio-ecological complexity of AQ communication. Challenges included involving underrepresented groups and assessing long-term impacts of interventions. The findings offer insights into how public engagement can be utilised to inform the design of an AQ communication technology to support protective and sustainable behaviours. Future research should explore ways to enhance inclusivity, refine message design for different user groups, and assess the impact of participatory AQ communication initiatives.

Generative AI tools have entered university life faster than institutions have been able to govern them, prompting policy responses that range from outright prohibition to largely unguided adoption. This paper examines the shift from restrictive to empowerment-oriented AI governance in higher education through a convergent parallel mixed-methods study comprising a structured survey (n = 71) and twenty-five semi-structured interviews across eleven academic disciplines at a private university in Bangladesh. The study documents patterns of AI adoption, stakeholder attitudes toward institutional policy, and barriers to responsible use. Participants situated within more empowerment-oriented institutional environments, including those characterised by AI literacy support, clearer ethical guidance, and stronger faculty engagement, reported greater confidence in using AI responsibly and clearer understanding of acceptable practices. By contrast, participants operating under restriction-only policies more often described uncertainty, confusion, and rule evasion. Regression analysis further showed that AI familiarity, frequency of use, and policy awareness were significantly associated with stronger support for empowerment-oriented governance. These findings inform a five-pillar framework for responsible AI integration encompassing AI Literacy Integration, Stage-Based Access, Transparent Use Norms, Assessment Innovation, and Faculty Development. Informed by stakeholder evidence and refined in dialogue with existing literature, the framework offers a practical model for institutions navigating AI governance in resource-constrained contexts. The paper contributes empirical evidence from a developing-country setting that remains underrepresented in current scholarship and highlights how responsible AI governance can support more equitable and context-sensitive higher education.

While global debates on Artificial Intelligence (AI) often focus on existential risks, a more immediate concern for developing economies is that uneven AI adoption will deepen existing regional inequalities. This paper presents a state-level empirical analysis of AI readiness disparities within India, quantifying how structural preparedness and public interest in AI co-vary across 28 states and the National Capital Territory of Delhi. We construct an AI Readiness Composite Score (AIRCS) that integrates educational outcomes, economic capacity, and digital infrastructure, and relate it to a normalised AI interest index derived from Google Trends data for AI-related search terms. The results show a positive but far from deterministic association between readiness and interest. Building on this joint distribution, we identify four readiness–interest profiles: high readiness and high interest leaders, high interest and medium readiness aspirants, low readiness and low interest priority states, and a large band of medium readiness and medium interest regions that track the national average relationship. This typology reveals that AI-driven growth is likely to follow a highly uneven spatial pattern, in which a small number of states are positioned to capture disproportionate benefits while others risk long-term exclusion. We argue that national AI strategies premised on universal digitalisation are unlikely to correct these imbalances. Instead, we propose a contextual integration framework in which AI policies, investments, and applications are tailored to the specific readiness and interest profiles of different state clusters, with the goal of preventing a new, AI-intensive layer of the digital divide from emerging within India.

Sociotechnical systems, such as language technologies, frequently exhibit identity-based biases. These biases exacerbate the experiences of historically marginalized communities and remain understudied in low-resource contexts. While models and datasets specific to a language or with multilingual support are commonly recommended to address these biases, this paper empirically tests the effectiveness of such approaches for gender, religion, and nationality-based identities in Bengali, a widely spoken but low-resourced language. We conducted an algorithmic audit of sentiment analysis models built on mBERT and BanglaBERT, which were fine-tuned using all Bengali sentiment analysis (BSA) datasets from Google Dataset Search. Our analyses showed that BSA models exhibit biases across different identity categories despite having similar semantic content and structure. We also examined the inconsistencies and uncertainties arising from combining pre-trained models and datasets created by individuals from diverse demographic backgrounds. We connected these findings to the broader discussions on epistemic injustice, AI alignment, and methodological decisions in algorithmic audits.

Recent geographic shifts in high-tech labor markets have redirected attention toward economic growth in small cities and rural regions across the United States. We examine how this shift has spurred economic development organizations, think tanks, and rural municipalities to leverage digital data in local economies as a strategic tool for attracting investment. Drawing on long-term ethnographic research on rural development policy and practice, we turn to the design of a new type of development tool: the digital asset map. In this paper, we describe how nationwide mapping and data visualization initiatives push for the digitization of rural community-based asset data, from broadband infrastructure to breweries. While these systems are intended to promote sustainable rural development, and to help leaders in their efforts to repopulate and revitalize rural communities in decline, their design often encourages comparison across communities in ways that reify divides between the rural “haves” and “have-nots”. By privileging quantifiable data, digital asset maps risk marginalizing communities whose strengths are less easily captured in datasets, potentially reinforcing existing patterns of uneven rural development.

Civic technology platforms are increasingly developed through collaborative processes involving diverse stakeholders. Yet existing participatory design literature offers limited theoretical attention to collaborative trajectories that deviate from a canonical model in which co-design is structured and intentional from the outset. This paper presents a reflective account of the development of “dédédé”, an online platform that invites citizens to informally document and share observations about their neighborhoods. The platform was initially conceived and developed by a team of technologists, then evolved through two co-design phases involving progressively broader stakeholder communities: first with urban designers and other practitioners experienced in facilitating machiaruki (“town-walking”) workshops, then with city officials and real estate developers interested in leveraging the platform for machizukuri (“town-building”) efforts. We show that these phases produced qualitatively distinct collaborative dynamics: the first characterized by receptive redesign, in which domain expertise corrected the implicit assumptions of the technical team; the second by active negotiation and resistance, in which the design knowledge accumulated in the first phase became a resource (design anchor) for mediating competing stakeholder visions. We argue that such staged co-design processes represent a recognizable and undertheorized mode of civic technology development, with distinct benefits and risks.

The rapid evolution of artificial intelligence and large language models have created a new dynamic of student engagement beyond academic contexts. While prior research focused on AI’s role in providing academic and emotional support, less is known about the motivations to personalize AI, its diverse usage, and the intricate relationship between AI and the student population, especially in the context of the Global South. Our study fills this gap with findings drawn from 21 semi-structured interviews with undergraduate students from diverse backgrounds in Bangladesh, revealing usage as companionship, intimate partnership, worldbuilding, reminiscence of dead family members, and prioritization of AI over human relationships. These findings shed light accounting the local cultural norms on shaping AI adoption in Bangladesh, highlighting both opportunities and risks, and provide essential insights to inform necessary guidelines. We further connect our findings to the broader issues in Human-AI interaction and offer future research directions.

Language operates as a mechanism of both marginalization and resistance, especially for minority communities navigating insensitive and harmful speech online. As content moderation increasingly depends on large language models (LLMs), concerns arise about whether these systems can recognize culturally insensitive speech–language that disregards or marginalizes the cultural and religious perspectives of historically underrepresented communities, often through implicit erasure, misrepresentation, or normative framing, rather than overt hostility. Focusing on Bangladesh’s Hindu and Chakma communities – the country’s largest religious and Indigenous ethnic minorities, respectively – this paper investigates the epistemic limits of LLM-based moderation systems and explores methods for incorporating minority perspectives. We co-created a culturally grounded corpus of insensitive speech with community members and integrated their narratives into moderation pipelines using retrieval augmented generation (RAG). Our tool, Mod-Guide, improves LLM sensitivity to minority viewpoints by leveraging contextual cues derived from lived experience. Through mixed-method evaluations involving both minority and majority participants, we demonstrate that RAG-enhanced moderation responses are more contextually accurate and perceived differently across ethnic lines. This work advances research in human-computer interaction, AI ethics, and social computing by foregrounding restorative justice and hermeneutical inclusion in the design of content moderation systems.

Indigenous children in Bangladesh’s Chittagong Hill Tracts (CHT) face persistent barriers to primary education, including language mismatch, economic constraints, and limited infrastructure. Prior work often frames these challenges as problems of access; however, we argue that educational inequality in this context is fundamentally mediated through social and institutional structures.

We present a four-phase mixed-methods study (2021–2026) involving 146 participants across five stakeholder groups. Our findings show that learning is shaped not only by technology availability but by networks of intermediaries, such as teachers and community members—who mediate access, interpretation, and trust. We further identify how language mismatch contributes to gradual disengagement from classroom participation, and how uneven access to devices produces intermediary-dependent learning structures. Based on these insights, we contribute (1) a multi-phase, corroborative, multi-actor study and dataset on indigenous primary education in Bangladesh; (2) the concept of intermediary-mediated learning systems; (3) a theoretical account of gradual disengagement as epistemic exclusion; and (4) a distributed user model with system-level design principles for mediated learning contexts.

In Bangladesh, "photo-cards", standardized news graphics widely shared on social media, have become a dominant format for news dissemination, but their visual authority is increasingly exploited to circulate misleading claims. This study examines how manipulated photo-cards imitate the branding of trusted news organizations and how audiences interpret these artifacts in online discussions. We conduct a qualitative analysis of 167 fake photo-cards collected from Facebook between November 2025 and March 2026, focusing on 20 artifacts that closely mimic the visual identity of major Bangladeshi news broadcasters and analyzing approximately 400 associated user comments. Our findings show that these artifacts replicate familiar layout and branding cues, allowing fabricated claims to borrow the visual credibility of established media institutions. Comment discourse reveals that most users initially react to the claim rather than questioning authenticity, while only a small fraction attempt correction. These results highlight how visual familiarity, cognitive heuristics, and social interaction shape the interpretation of misinformation in mobile-first environments.

Artificial Intelligence (AI) readiness in the Global South is often framed as a matter of infrastructure gaps or policy ambition. Drawing on a multi-method qualitative study of 35 university programs, 59 stakeholder interviews, and curriculum benchmarking in Bangladesh, we reconceptualize AI readiness as a sociotechnical condition shaped by material infrastructures, human capacity, and curricular governance. Our findings reveal how GPU scarcity, limited faculty upskilling, opaque mentorship networks, entrenched gender disparities, and the near absence of Responsible AI instruction collectively constrain institutional capacity. Using concepts from Science and Technology Studies, including blackboxing, nested infrastructures, strain, and path dependency, we show that readiness deficits are not isolated shortcomings but outcomes of layered bureaucratic systems, historical lock-in, and global market dependencies. We further argue that AI education in Bangladesh is shaped by postcolonial dynamics that privilege global labor alignment over locally grounded innovation. We contribute an infrastructural reframing of AI readiness for HCI and ICTD, empirical evidence of institutional and epistemic barriers in a Global South context, and actionable design and policy pathways for building equitable, human-centered AI ecosystems.

Low-resource and tribal languages in India are at acute risk of digital extinction because contemporary language technologies overwhelmingly target high-resource languages and centralized, cloud-hosted large models that are expensive to train, deploy, and adapt. Centralized automatic speech recognition (ASR) and text-to-speech (TTS) pipelines require large amounts of labelled data, stable connectivity, and significant compute resources, all of which are misaligned with the infrastructural realities and socio-economic conditions of many tribal communities in India. These constraints are particularly visible for the Soliga community, a Dravidian-language-speaking tribe in the Biligiri Rangaswamy Hills of Karnataka, whose language has no widely used script and for which digital resources remain scarce. We present a socio-technical framework and prototype for building low-cost voice models with and for the Soliga tribe using offline, edge-based federated learning. The framework combines community-governed data collection and consent processes, on-device training of compact ASR and keyword-spotting models, and an intermittent-connectivity aggregation protocol that runs on low-cost edge hardware. Our experiments show that, even with only 5 hours of Soliga speech, a centralized Wav2Vec 2.0 baseline achieves a word error rate (WER) of 37.95% and character error rate (CER) of 11.11%, and that a federated frugal architecture trained on the same small corpus can approach this performance (WER 44.20%, CER 15.80%) while keeping data local and under community control.

Social media platforms increasingly function as informal infrastructures to seek justice in contexts where formal justice systems are inaccessible or distrusted. We examine how Bangladeshi individuals leverage Facebook to disclose personal experiences of harm (e.g., sexual harassment, fraud, and abuse) and pursue justice. Our in-depth interviews with 18 participants unpacked the motivations, strategic choices, and sociotechnical risks involved in such disclosures. Our participants carefully navigated trade-offs between visibility, credibility, privacy, and legal vulnerability. Our findings contribute to HCI by (1) conceptualizing justice-seeking behaviors in low-trust, high-risk environments; (2) offering empirical insights into post-curation, audience targeting, and identity management; and (3) identifying design opportunities for safer, more equitable platform interactions. We present implications for platform governance and moderation, including features that support credibility, mitigate emotional harm, and enhance the legibility of justice claims. This work advances critical HCI and social computing perspectives on safety, care, and power in digital participation.

Rapid digitization across government services, financial platforms, and telecommunications has intensified the collection and processing of large-scale personal data in Bangladesh. In response, the state has introduced multiple regulatory instruments, including the Personal Data Protection Ordinance, the Cyber Security Ordinance, and the National Data Governance Ordinance in 2025. While these initiatives signal an emerging legal regime for data protection, little scholarly work examines how they operate collectively in practice. This paper presents a legal-institutional analysis of Bangladesh’s data protection regime through a systematic review of these three ordinances, mapping how they collectively structure personal data governance and where frictions arise in their translation into institutional and regulatory enforcement. Our findings reveal that this emerging regime is constrained by uneven regulatory capacity, limited institutional independence, and a misaligned legal assumption of individualized, autonomous data subjects. We further demonstrate how these frameworks invisibilize prevalent sociotechnical layers, such as mediated access via “human bridges” and informal data flows, rendering formal protections difficult to operationalize. This paper contributes to HCI scholarship by expanding the concept of data protection as a complex sociotechnical design problem shaped by the informal infrastructures of the Global South.

In Bangladesh’s rapidly expanding informal e-market, small-scale sellers use social media platforms like Facebook to run businesses outside formal infrastructures. These sellers rely heavily on platform algorithms—not just for visibility, but as active collaborators in business operations. Drawing on 41 in-depth interviews with sellers, buyers, and stakeholders, this paper examines how people in informal e-market perceive and interact with the algorithm as a “team member" that performs sales, marketing, and customer engagement tasks. We found that while sellers and local tech entrepreneurs are intrigued to develop services to support this industry, buyers and investors of the industry put their greater trust in human interactions. This surfaces a postcolonial tension associated with cultural values, local tech education and training, and a mismatch between the global and Bangladeshi e-markets’ growth. We expand this discussion from multiple ongoing HCI, political design, and AI design angles. We also address the postcolonial tension and support the decoloniality movement in informal e-markets by proposing the DAIEM framework that consists of six components: autonomy and agency; resistance; locality, culture, and history; rationality; materiality; and advocacy. Supporting the decoloniality and informality sentiment in informal e-market and other similar sectors, DAIEM will serve both as a guideline for algorithm design and as an analytical tool.