The analysis focused on two main areas: workflow efficiency and the feasibility of integrating the solution into the standard design process of an architectural firm. Particular attention was paid to the integration of executable scripts with digital engineering knowledge resources, encompassing national legal regulations, technical standards, and construction technology issues.
The system was implemented on the Antigravity platform and is based on two main components: knowledge bases and automated tools. Knowledge bases serve as structured repositories for source data. The test environment utilized datasets comprising Polish construction-related legal acts, including the provisions of the Building Law and Technical Conditions. The documents were converted to Markdown format, which facilitates their searching and processing by language models. In more complex interpretative cases, the system utilizes the MCP protocol to supplement information via external resources, e.g., current case law.
The database was also expanded to include literature on general building construction, structural engineering, and building physics. A crucial element of the implementation was the database of design details. Raster (JPG) and vector (DXF) files were linked using a unified descriptive layer in Markdown format. This approach enables the simultaneous analysis of the technical description, geometry, and graphic material.
The second system component is an automated tool that transfers details from the database directly into the Autodesk Revit environment. The operational logic was developed using a high-compute model (Gemini 3.1 Pro/High), whereas lighter models (e.g., Gemini Flash) are applied for repetitive operations, thereby reducing computational costs. The actual process of converting geometry from DXF to Revit’s native format is executed by a pre-written Python script. Utilizing constant code rather than generating it on the fly ensures greater operational stability and reproducibility of results.
Test results indicate that the solution’s effectiveness depends primarily on the quality of the input data preparation, specifically the vector files and the structure of the knowledge base. Proper formulation of technical descriptions is also crucial—they should focus on the operational principles of the solutions and their execution correctness, rather than solely on implementation specifics. A properly prepared database can be gradually expanded with further elements, such as technological guidelines or product data.
The applied approach allows for a direct connection between the design environment and the knowledge repository. A key feature of the solution is its independence from any specific language model—the system can interact with various models depending on the needs. In practice, this signifies a transition from the passive use of AI tools to a more active working model, where an agent operates simultaneously across multiple data sources and interacts directly with the design software.
Creating custom knowledge bases is of critical importance in this context. It leverages organizational knowledge, mitigates the risk of generating imprecise responses, and standardizes design protocols. The solution can also be utilized for real-time project verification—e.g., through screenshot analysis and compliance assessment with current regulations.
This approach can be further developed to include, among others, the automated generation of technical documentation (e.g., MasterFormat/NBS Specifications), the drafting of project descriptions based on geometric models, and the verification of building parameters against legal and technical requirements.
Disclaimer
The reference materials (including design details and drawings) shown in the attached video and utilized in the test knowledge base remain the intellectual property of their respective authors. They have been used under the right of quotation (fair use), solely for educational, analytical, and demonstration purposes as part of technological verification and a Proof of Concept. The knowledge base built upon them serves strictly as a private testing environment and is not subject to any form of reproduction, distribution, or commercialization.