September 6, 2026
In the construction and design industries, BIM and other digital technologies are often discussed in the context of “centralizing information.”
In practice, however, many workplaces face a more fundamental problem before information can even be centralized: no one clearly knows where existing documents are stored, and simply locating files from previous projects can take a considerable amount of time.
It is common to be asked to “create a drawing based on a similar project we handled in the past,” while finding that the previous project data itself is surprisingly difficult to locate.
In this article, I describe a preliminary approach to this problem: creating a JSON-based index of the entire folder and file structure of a shared directory, not limited to CAD or BIM data.
The objective is to create a foundation for future information organization, searching, and AI-assisted processing.
1. Creating a Local Tool to Convert a Folder Hierarchy into JSON
The first step is to understand the current structure of shared folders and project directories.
For this purpose, a simple tool can be created that reads the folder and file hierarchy of a selected directory and exports it as a JSON index.
For example, a generative AI can be instructed as follows:
Create a locally running HTML tool that reads the folder and file hierarchy of a selected directory and generates an index in JSON format.
The tool must operate in read-only mode and must not modify, overwrite, move, or delete any original folders or files.
It should also include a progress indicator so that the user can monitor the scanning status when processing large directory structures.
Using such a tool, the structure of an entire shared folder can be stored as JSON data that can later be processed by AI or other search and classification tools.
The important point is that there is no need to reorganize all existing CAD, BIM, and project data from the beginning. The first goal is simply to make it possible to identify
what folders and files currently exist and where they are located.
2. Creating a Specification Classification Database in JSON
Once the folder hierarchy JSON has been created, the next step is to import that data and generate another JSON file organized by project specifications or other useful classification criteria.
The hierarchy JSON is suitable for representing the existing folder structure as it is. However, when the objective is to search for projects that previously used a particular specification, drawings created under similar conditions, or reference documents from comparable past projects, a simplified database-oriented JSON structure is often easier to search.
Such a structure is also more suitable for importing into generative AI systems for classification, filtering, and analysis.
A practical workflow is to first upload the hierarchy JSON to an AI system and ask it to examine the data and suggest:
- which attributes should be used for specification-based classification,
- how much information can be inferred from folder and file names, and
- what JSON structure would be appropriate for a searchable database.
After determining the classification structure, the AI can then be asked to create another local HTML tool.
Create a locally running HTML tool that imports a JSON file containing a folder hierarchy and generates a separate database JSON file classified by specification.
The original hierarchy JSON must remain unchanged, and the classification results should be exported as a separate JSON file.
By maintaining two separate datasets —
a hierarchy JSON representing the actual folder structure and
a classification database JSON optimized for searching — it becomes possible to introduce a searchable information layer without significantly changing the existing shared-folder environment.
Centralizing information through BIM and other systems remains an important long-term objective. However, before attempting full information centralization, there is also considerable practical value in making the large amount of existing project data and historical documentation
searchable in the first place.
Any classifications inferred by AI from folder or file names are not the source data itself. For practical use, the results should be reviewed and treated as search-support data rather than as a final technical judgment.
