| El Faro: An Advanced Frugal AI Stack for Context-Driven Resilience | Jose Daniel Ruiz | 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. |
| Spondon: A Low-Cost, Multi-Sensor Wearable for Understanding Stress in Resource-Constrained Countries | Shahariar Ifti; Rahat Jahangir Rony; Shuvashish Chakraborty; Asif Mahbub; Nova Ahmed | 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. |
| Sovereign by Design: A Service Architecture for Accountable Small-Language-Model Deployment in Citizen-Facing Government Services | Venumadhav Venkata Kalluru | 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. |
| Designing Sode: A Culturally Grounded Web Application for Social-Emotional Learning Among Ghanaian Preschoolers | Angela Yaa Kaare Benning; Kwabena Ampadu Bamfo | 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. |
| Who Decides What Art Means? Interpretive Authority, Cultural Equity, and Conversational Access for Blind and Low-Vision Museum Visitors | Nisheta Gupta; Arnav Sharma; Atharva Nayak | 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. |
| Designing Voice-Based QR Payment Confirmation for Small Vendors: A Formative Pilot Study | Shreya Ganeju; Anubhav Subedi; Ayush Kaji Dangol; Sabin Shrestha; Bidhan Dev; Supriya Khadka | 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. |
| Towards Sustainable Waste Management in Nepal: A Low-Cost Edge-AI Prototype for Waste Classification | Samikshya Dhamala; Raunak Regmi; Ayush Kaji Dangol; Unika Ghimire; Shreya Bhatta; Supriya Khadka | 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. |
| The Doomscrolling Dilemma: How Multilingual Short-Form Videos Reshape the Attention of University Students and the Decline of "Adda" in Bangladesh | Ehsanul Haque Ome; Sadar Ahmed; Mumtahina Mahbub Chowdhury; Umme Jannat Taposhi; Farida Chowdhury; Marshia Nujhat | 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. |
| From Power-On to Personalization: Understanding Smartphone Setup Experiences in Bangladesh. | Sabiha Ishrat; Imran Bhuiyan Dhiman | 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. |
| The Limits of Technological Disruption: AI Imaginaries for Parkinson's Care in Low-Resourced Public Health Systems | Luis Ramos-Pozo; Juan Pisco-Jordán; Gabriel Madroñero-Pachajoa; Gonzalo Gabriel Méndez; Javier Tibau; Marisol Wong-Villacres | 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. |
| ClimateSignal Logger: Co-Designing a Community-Controlled Environmental Data Logger for Heat Justice in Little Haiti, Miami | Oluwafemi Oladosu; Keith Maull; Agbeli Ameko; Curtis Walker; Amy Quarkume | 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. |
| Designing the Digital Gateway: A Conceptual Framework for a Digital Platform to Empower Conservative Women Entrepreneurs of Bangladesh | Md Fardin Alam; Zarin Tasnim; Arani Annesha; Sabiha Ishrat; Mashiyat Mahjabin Eshita; Farida Chowdhury | 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. |
| A Tool of One’s Own: Motifs, Abduction and Non-Generative World-Building | Meral Senturk; Umut Tasa | 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. |
| Loving Baby: An AI-Supported Robotic Companion for Primary Education -- Early Evidence on Emotions, Engagement and Well‑Being | Ka Yan Fung; Kuen Fung Sin | 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. |
| Monitoring-as-Practice: An Exploratory Study of Everyday Air-Quality Monitoring Through Practice Theory for WAQMs | Wensi Cai; Daria Morozova; Laura Pruszko; Mike Mannion | 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. |
| Environmental Constraints in UAV-Based Litter Detection: Deployment Insights from a Forest Park Case Study | Jason Zhao | 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. |
| Repetition is not Redundancy: AI and Relational Listening in Communicating about Extreme Weather | Madeleine Antonellos; Angelina Aquino; Ian Mongunu Gumbula; Jens Cheung; Nicola J Bidwell | 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. |
| Are You Willing to Wait? An SMS Platform to Measure the Value of Time in the Field | S. Reid Dickerson; Alice Duhaut; Geetika Nagpal; Nick Tsivanidis; Daniel Bjorkegren | 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. |
| Toward Faith-Aligned Mental Health Chatbots: Understanding Muslim Users' Expectations and Design Needs | Mohammad Rakin Uddin; Afsana Hossain Anima; S M Rhydh Arnab; Susmita Biswas; Jannatun Noor | 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. |
| “There Are Things I Can’t Ask People”: User Perceptions of LLM-Based Chatbot for Sexual and Reproductive Health in India | Sara Moin; Himani; Pushpendra Singh | 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. |
| “We Don’t Know Where to Start”: Exploring Designers’ Perspectives on Accessibility in Voice User Interfaces Design | Adesayo Justina Adefuye; Laura Pruszko | 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. |
| When Tutorials Become Infrastructure: Bangla Developer Vlogs, Learning Access, and Precarious Digital Labor in Bangladesh | Jannatun Noor; Susmita Biswas; Atiqur Rahman; Jannatul Feardous Nafsi; Mst. Dilruba Khanom Dolon; Md Farhan Jahin | 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. |
| “Spilling PII into the Lake:” Design and Evaluation of LLM-based Redaction for Community Privacy in Grassroots Volunteer Organizations | Rudra Prakash Singh; Esther Han Beol Jang; Paul Philion; Kurtis Heimerl | 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. |
| A Study of Usability and Accessibility of Private and Public Hospital Websites in Bangladesh | Maisha Chowdhury Neha; Misha Mahenur Alam; Marshia Nujhat; Umme Jannat Taposhi; Farida Chowdhury | 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. |
COMPASS