Joseba Estévez is a social anthropologist and Research Assistant Professor at the Hong Kong Institute for the Humanities and Social Sciences, The University of Hong Kong.
x
x
x
x
x
x
x
x
Sacred Hong Kong connects maps, walks, temple registers and heritage records to explore religious landscapes. The methods and sources below show where records come from, how they are used, and what they can or cannot say; figures describe the dataset, not the present state of any congregation.
This project is being developed through an integrated workflow linking Perplexity AI, Manus AI, GitHub, and Vercel. This approach connects preliminary research and interface design with data processing, software development, version control, and deployment. Each platform serves a distinct function, allowing the prototype to evolve through successive cycles of research, implementation, and revision.
Perplexity supports the exploratory research and planning phase by identifying relevant web resources, examining interface layouts, and comparing approaches to navigation and content presentation. These findings inform the project’s design specifications, with particular attention to the organizations of research material and its accessibility to different audiences.
Manus supports implementation by translating these specifications into data-processing scripts and responsive interface components within a virtual development environment. Its role includes cleaning and structuring source data, generating code, and updating the project repository. AI-assisted development facilitates prototyping, while the accuracy of the content and the suitability of design decisions remains subject to human review.
GitHub provides the central repository for the application’s source code and a record of successive modifications. Version control enables changes to be traced, compared, and revised, supporting continuity across development stages and making the implementation process more transparent.
Vercel provides the deployment and hosting infrastructure. With the repository integration configured, code updates initiate the application’s build and deployment process, making revised versions available online. This arrangement connects ongoing development with web delivery while reducing dependence on a locally configured production environment.
Together, these tools support an iterative approach to building the prototype. Automation reduces repetitive technical work and shortened the transition from source material to a functioning interface. Project design, AI orchestration, research interpretation, source verification, ethical decisions, and the evaluation of the resulting application remain human responsibilities.
2,657 points with bilingual names, religion, denomination, branch, language and Chinese address. Published on the Purdue University Research Repository under CC0 1.0. Its strength is coverage of venues without buildings: upper-floor units, wall shrines, yoga studios classed as Hindu. Its weaknesses are geocoding slips (five corrected on the trails), duplicates and single-religion labels for syncretic sites.
Declared monuments and graded buildings as polygons with grade, year, bilingual names and an image-list endpoint. Joined by containment: 69 OSAGE sites fall inside 55 monuments or graded buildings. Conversely, 288 of the register's 321 religious-looking entries have no OSAGE point within them, most in Yuen Long, North and Tai Po, where village ancestral halls and study halls are heritage but not, by the atlas's criteria, active religious venues.
Landmarks were matched to Wikidata items by name search, yielding coordinates, inception dates, heritage status, Wikipedia links and a Commons image with author and licence. Hong Kong-wide SPARQL queries over place-of-worship classes were unreliable at the public endpoint, so a production build would use a Wikidata dump or the Commons geosearch API with name filtering. 33 of the 38 trail stops carry a licensed image.
Coastlines of 1845, 1888, 1904 and 1924 redrawn by the Hong Kong Historic Maps project from Admiralty charts and government surveys, with a 2004 coast for comparison. Only stretches more than 40 m from the 2004 coast are kept, which reduces 2.7 MB of lines to the reclaimed sea alone. Tin Hau temples were placed from the register's geocodes first, then OpenStreetMap, then the atlas; 58 of 124 listed temples could be placed with confidence, and the shore distance is a straight line to the nearest coast segment in each year.
1,492 places of worship were read from OpenStreetMap through the Overpass API and used to refine positions on Lantau and for Tin Hau temples; 90 temple nodes on Lantau are shown as a third witness. The 2021 Population Census small-area tables (Street Blocks and Tertiary Planning Unit Groups) and their boundaries come from the Census and Statistics Department through the Common Spatial Data Infrastructure portal; every atlas point was assigned to its census area by point-in-polygon, and rates use resident population as the denominator.
Deity birthdays follow the Chinese Temples Committee's festival calendar, supplemented from the Chinese Wikipedia article on deity birthdays (shendan) for deities the Committee does not list. A temple is counted on a day if the deity appears in its register entry; the darker bar counts it only when that deity is named first. The 48 Jiao come from the Chinese Wikipedia list of Hong Kong Taiping Qingjiao and the ICH inventory; village alliances and islands were placed at settlement centroids by hand, so positions are indicative to a few hundred metres.