This paper proposes a web-based GIS framework for “caring maps” in post-disaster cities, using the Turkish city of Antakya (Hatay) and Beirut as comparative case studies. It asks howinteractive mapping can simultaneously function as: (1) a community-facing interface for resilience and everyday recovery, (2) a digital archive of fragile heritage and lived experience, and (3) a site where AI-driven automation is carefully integrated without erasing the situated, small-scale nature of the data.The first case, Antakya, is a historically multi-religious and multi-ethnic city in southern Türkiye that was heavily damaged in the February 2023 earthquakes. Many historic neighbourhoods, religious buildings, vernacular houses, family gardens and micro-economies were partially or entirely destroyed. Building on ongoing fieldwork in visual anthropology and digital heritage, I have been developing a prototype web-GIS that combines geolocated points, lines and polygons with 3D models, photographs and narrative annotations. The map includes not only “major” monuments but also small, intimate sites: a now-collapsed family home, a corner shop that became a meeting point after the disaster, or a temporary container settlement where religious rituals have been reconfigured. Data sources include GPS traces collected during walks, drone and terrestrial photogrammetry, structured interviews and informal conversations, as well as archival materials where available. The dataset is intentionally small, curated and deeply contextual, rather than exhaustive or statistically representative, aligning with digital humanitieswork on “small data” as processual and interpretive rather than purely quantitative (Ciula et al. 2021).The second case, Beirut, considers digital mapping efforts in the wake of the 4 August 2020 port explosion, a largely man-made disaster whose blast wave damaged large parts of the city and its historic neighbourhoods. Here the focus is not on a single map that I have built myself, but on a landscape of initiatives by local researchers, architects, activists and NGOs who have used mapping to trace damage, document buildings at risk, record testimonies and visualise displacement and reconstruction. Drawing on published reports, open datasets and collaborations with practitioners, the paper reconstructs how these mapping practices have tried to balance urgent documentation needs with questions of consent, visibility and the politics of who gets mapped and why, resonating with recent calls for critical, reflexive disaster cartography (Carraro, 2023). The juxtaposition of Antakya (earthquake, ongoing fieldwork) and Beirut (port blast, partly secondary analysis) allows us to ask how similar digital tools are enrolled into different temporalities of crisis, memory and political struggle. Methodologically, the paper is grounded in visual anthropology, critical GIS and design research. It combines hands-on prototyping of a web-based GIS application (in the Antakya case) with qualitative analysis of mapping practices (in the Beirut case). Rather than treatingdata” as a neutral substrate, the paper foregrounds the situatedness of each point, polygon and media attachment: who produced it, under what conditions, and for whom. This approach echoes work in digital humanities and heritage risk management that emphasises the interpretive and value-laden nature of spatial datasets, particularly in fragile environments (Yu et al. 2025).The technical backbone of the Antakya prototype is a browser-based map built with open-source mapping libraries. It can display multiple layers: base maps, pre- and post-disaster imagery, user-defined heritage layers and links to 3D content hosted on external viewers. 3D elements—such as photogrammetric models of buildings, courtyards or small religious interiors—can be opened from the map as pop-up viewers or side-by-side “before/after” panes, enabling users to move between cartographic overview and volumetric presence. Attribute tables are designed not just for building typologies or damage grades, but also for social and cultural descriptors: the communities that used the space, ritual practices attached to it and anecdotal stories, often captured as short text or audio.Within this framework, the paper explores how AI-assisted automation can be woven into the workflow without turning the project into a generic “AI platform”. Rather than training large bespoke models or outsourcing interpretation, AI is positioned as a set of modest tools that operate in the background and remain accountable to human researchers and community partners. Concretely, three areas are considered:

Translingual support and textual care.Field notes, interviews and community submissions frequently move between Turkish,Arabic and English (in Antakya), and between Arabic, French and English (in Beirut). AI-based machine translation and summarisation can help produce parallel descriptions and interface labels, lower the barrier for multilingual browsing and generate first-draft summaries of long narratives. At the same time, the paper stresses the importance of keeping original-language versions visible, flagging machine-generated text transparently and retaining human editorial control, in line with debates on trust, collaboration and shared professional ethics in AI-supported archival work (Jaillant and Rees 2023).

2. Semi-automated annotation of small visual datasets.In both cities, photographs of facades, interiors and street corners accumulate quickly, butmanual tagging is time-consuming. Lightweight computer vision tools can suggestpreliminary labels such as “crack”, “rubble”, “arch”, “dome” or “container housing”,which are then confirmed, corrected or rejected by the researcher and, where appropriate,by community partners. The aim is not to let AI “decide” what damage is, but to reduce repetitive labour and to surface images that might otherwise be forgotten in local folders. This is framed as annotation with small data, where each correction also becomes amoment of reflection on model bias and misrecognition (Ciula et al. 2021).

3. Ethical prompts and alerts in map use.

Finally, the paper sketches how rule-based and AI-augmented prompts might be used to slow down certain interactions. For example, when a user attempts to export coordinatesof a sensitive shelter location, or to download a bundle of images from a religious site, the interface could trigger a contextual message explaining why such data is delicate and asking the user to reflect on intended use. Here, automation does not optimise efficiencybut introduces micro-pauses for ethical engagement, paralleling discussions in critical GIS about repoliticising disaster-related maps (Carraro 2023). A central argument of the paper is that these AI components only make sense when the overall system is understood as a “caring map” rather than a neutral information platform. By caring map, I mean a cartographic assemblage that acknowledges loss and absence as much as presence; foregrounds relationships and responsibilities between mappers, mapped communities and future users; and treats technical choices (layer structure, default zoom levels, pop-up content) as ethical decisions. In Antakya, this might mean deciding not to publicly map some locations at all, or to keep certain narratives accessible only through mediated access in workshops. In Beirut, it might mean re-scaling damage maps that were originally designed for international advocacy so that they can also support local conversations about everyday life and long-term decisions about staying or leaving.

The comparative design of the paper allows us to reflect on differences between natural and man-made disasters without collapsing them into a single category. In Antakya, the earthquakes are framed both as geophysical events and as failures of regulation and governance; yet much public discourse still falls back on the language of “natural catastrophe”. In Beirut, the explosion is widely understood as a result of political negligence and corruption, layered on longer histories of civil war and economic crisis. By tracing how web-based maps encode or contest these framings—through their choice of basemaps, legend categories, filters and textual descriptions—the paper highlights the political work of digital cartography in post-disaster memory, complementing broader heritage-risk and resilience literature that uses GIS to analyse multi-hazard threats (Yang et al. 2023; Dammag et al. 2025).From a digital humanities perspective, the contribution is threefold. Conceptually, the paper proposes “caring maps” as a framework for thinking engagement in post-disaster GIS, bringingtogether community resilience, heritage studies and critical data ethics. Methodologically, it details a mixed-method workflow that combines ethnographic fieldwork, 3D reconstruction, web mapping and carefully delimited AI assistance, explicitly tailored to small, sensitive datasets (Ciula et al. 2021; Yu et al. 2025). Practically and design-wise, it offers concrete interface patterns—such as layered narratives, multi-language toggles and consent-aware download options—and discusses how they might be implemented in future iterations of the Antakya prototype and in analogous Beirut projects. In terms of format, the paper will include screenshots and simple diagrams of the Antakya web map prototype, schematic representations of AI-supported workflows (translation, tagging, alerts) and comparative vignettes from Beirut mapping initiatives. While the Antakya platform is still under active development, an early-stage version will be available online by the time of the conference, allowing participants to test ideas in a live environment. Throughout, the emphasis is not on presenting a finished “solution”, but on sharing an iterative, reflexive design process that foregrounds engagement with communities, languages and technologies in fragile urban landscapes. By placing Antakya and Beirut side by side, and by explicitly interrogating the roleof AI within small, situated datasets, the paper invites digital humanists to re-think what counts as “data”, “tool” and “impact” in post-disaster contexts. It argues that the future of digital heritage and GIS in such environments is less about scaling up to ever larger datasets and more about cultivating careful, accountable infrastructures that can support memory, critique and everyday forms of resilience over the long term.