SmartArch AI system flow

1. General Flow

This system is an AI-powered, end-to-end automated platform designed to facilitate the digital analysis of architectural plans. Upon logging in, users complete the payment transaction via the POS system, then upload their architectural plans (in PNG or JPG format) to initiate the analysis process. The system analyzes the uploaded plans by identifying, classifying, and evaluating the architectural and structural elements within them, subsequently presenting the results to the user through various output formats (DXF, PDF, PNG).

2. SYSTEM FLOW

2.1 Login → POS

The system commences with a secure user authentication mechanism. The user: • Logs into the system • Is redirected to the payment screen (Virtual POS) • Gains access to the analysis functionality upon successful payment (e.g., for 3 projects)

2.2 Electronic Invoice → Mail

Following payment, the system automatically: • Generates an electronic invoice • Sends the invoice to the user’s registered email address

2.3 PNG Upload (Plan Upload)

The user: • Uploads architectural plans in PNG or JPG format

3. EXTRACTION OF ARCHITECTURAL ELEMENTS FROM IMAGES

This constitutes the core component of the system. Detected elements include: • Door • Window • Wall • Axis & Axis Point • Stairs • Elevator Shaft The process involves: • Initialization using YOLO or a comparable object detection model • Establishment of relationships through a relational model • Calculation of confidence scores employing an Ensemble and Bayesian model For each element: • An identification number (ID) is assigned • Location is determined • Confidence score (%) is computed

4. CONCLUSION

Following analysis, the system: • Summarizes all results • Delivers a comprehensive and meaningful output to the user

5. LOW CONFIDENCE SCORE ELEMENTS

A key component in the workflow is represented by the statement: “Individuals below the threshold can be eliminated.” This indicates that: • Elements with low confidence scores • May be removed either automatically or manually by the user This feature: • Minimizes false positives • Enhances the overall quality of the analysis

6. RE-ANALYSIS

The user: • Removes incorrect elements • Restarts the analysis process This functionality is critical as it: • Establishes a human-AI hybrid system • Facilitates continuous improvement

7. OUTPUTS

The system generates outputs in the following formats:

8.1 DXF

• CAD compatible • Axes represented as single lines • Columns and walls organized in separate layers

8.2 PDF

• Comprehensive report • Visualization of results

8.3 PNG

• Processed architectural plan

8.4 ALL PACKAGE

• All output files provided in a ZIP archive

This explanation is displayed automatically according to the language selected at entry.

SmartArch AI interactive plan editing flow

1. FLOWCHART: PNG/JPG PLAN UPLOAD AND ARCHITECTURAL ELEMENT ADDITION SYSTEM

SmartArch AI interactive architectural element editing flow

This diagram illustrates the workflow of an interactive system that not only performs automatic analysis of architectural plans but also permits direct user editing. Unlike traditional AI-based architectural analysis methods, this system does not operate solely on a "detect and produce results" basis. Instead, it enables users to actively engage with the plan by editing existing elements and adding new architectural components. The process depicted in the diagram comprises three primary components: 1. PNG/JPG plan upload 2. Integration of architectural elements into the image 3. User-assisted editing and development Through this approach, the system evolves from a simple analytical tool into an interactive architectural design platform. Users are provided not only with AI-generated results on the uploaded plan but also with the capability to control, modify, delete, and incorporate new elements as required.

2. PLAN UPLOAD: PNG / JPG

At this stage, the user progresses beyond merely uploading a file and gains access to an architectural plan workspace within the system. The plan thereby becomes a dynamic and interactive surface for user engagement.

3. AUTOMATIC + MANUAL HYBRID SYSTEM

The system’s primary strength lies in its integration of artificial intelligence with user intervention within a unified workflow. In other words, it is not a closed system relying exclusively on fully automated decisions; instead, it actively involves the user throughout the process. This approach functions as follows: • The system performs an automatic analysis of the plan. • It detects architectural elements such as walls, doors, and windows. • Identified elements are displayed on the plan. • The user reviews these elements. • The user can delete any inaccurately detected elements. • The user can manually add any missing elements. • The user can create new walls, doors, or windows as necessary. This framework is founded on the principle of AI and user collaboration. Artificial intelligence delivers a rapid, automated initial analysis, while the user refines this analysis through architectural expertise and visual verification. Consequently, the system operates both efficiently and with precision. Given that AI-generated results alone may not always be sufficient—since some architectural lines may appear similar (e.g., wall lines versus axis lines), door openings vary by drawing style, and window symbols differ between projects—user intervention significantly enhances the reliability of the outcomes.

4. INTEGRATION OF ARCHITECTURAL ELEMENTS INTO THE IMAGE

The term "Adding New Elements to the Architectural Plan" featured in the central section of the diagram denotes the system’s core functionality. At this stage, architectural elements are incorporated, modified, or refined on the uploaded PNG/JPG plan image. This can be achieved through two distinct methods: 4.1 Automatic Integration The AI model analyzes the plan and identifies architectural elements within the image. For example, it: • Recognizes existing walls. • Detects door openings. • Identifies window drawings. • Extracts the basic spatial configuration of the plan. These elements are processed as layers on the image, enabling users to clearly visualize the location of each architectural component on the plan. 4.2 Manual Integration Users have the capability to add new elements manually, which represents a critical feature of the system. This functionality ensures that users are not confined to AI-generated results but can tailor the plan according to specific requirements. Users can: • Add new walls. • Introduce doors on existing walls. • Insert windows on existing walls. • Remove incorrectly added elements. • Edit existing elements. • Preview changes before finalizing decisions. This feature enables the system to serve not only analytical purposes but also design and correction functions effectively.

5. TRIAL-AND-ERROR MECHANISM

This system does not require the user to make a definitive decision in a single attempt. Users have the ability to experiment with the plan, review the results, remove elements if unsatisfactory, and reintroduce them as needed. The trial-and-error process operates as follows: 1. The user adds an element. 2. The system displays the element within the plan. 3. The user evaluates the visual outcome. 4. If the element is unsuitable, it is deleted. 5. The element is re-added in a different location.

6. This process is repeated until the desired result is achieved.

This methodology is particularly critical for architectural plans, as determining the precise placement of doors or windows may not be straightforward on the first attempt. By exploring multiple configurations, users can identify the most appropriate layout. The trial-and-error system offers several advantages: • It empowers the user with freedom. • It ensures that mistakes are not permanent. • It facilitates iterative design. • It supports achieving the most accurate outcome. • It promotes confident plan development by the user. This feature enhances the system’s flexibility and user-friendliness. 6. USER + AI COLLABORATIVE PLAN The resulting plan is not merely an output automatically generated by artificial intelligence; it is also refined through user input. The plan is composed of three fundamental layers: 1. Original plan image 2. Architectural elements identified by artificial intelligence 3. Elements added or modified by the user This layered structure enables systematic management of each modification. For example, a wall added by the user can be maintained in a separate layer, while doors identified by artificial intelligence can be recorded as a distinct resource. Consequently, it is possible to track which elements originated from the system and which were introduced by the user. This tracking mechanism is essential for future reporting, undo operations, re-analysis, and output generation.

7. CONCLUSION

The system depicted in this diagram is not a simple artificial intelligence tool that merely reads and analyzes architectural plans. Rather, it is an interactive architectural design platform that places the user at the center of the process, allowing direct manipulation of the plan and enabling the attainment of the most accurate result through trial and error. The system: • Performs more than analysis alone. • Engages the user actively throughout the process. • Permits the addition of new walls. • Allows for the insertion of doors or windows into existing walls. • Provides the capability to delete and redo incorrect actions. • Integrates artificial intelligence with human oversight. • Delivers a more precise and optimized plan. This approach represents a significant advancement over traditional AI systems, as the user is not merely a passive recipient of results but an active participant who develops, refines, and directs the plan.

This explanation is displayed according to the selected site language for the second flow diagram.

SmartArch AI conclusion, filtering and download flow

3. Flow: Conclusion, Filtering and Download Workflow

SmartArch conclusion filtering and download flow

The objective of this section is not to accept the initial results produced by artificial intelligence as final; rather, it is to enable the user to visually review these results, verify them against a list, eliminate low-accuracy candidates, re-analyze using the remaining detections if necessary, and ultimately download DXF/PDF/PNG files.

1. Main Logic After Analysis

Upon analyzing an architectural plan, the system generates numerous candidate detections. These candidates may include doors, windows, walls, axis points, columns, curtain walls, stairs, elevator shafts, or other architectural structural elements. However, not every detection produced by artificial intelligence models can be considered definitively accurate. Therefore, the system provides the user with a control screen following the analysis. This control screen comprises two primary areas: • Visual Results List: Visual results displayed directly on the plan. • Detection List: A tabular compilation of candidate elements identified by artificial intelligence. This configuration allows the user to view not only the image but also the ID, type, confidence score, coordinates, size, and source of each candidate.

2. Visual Results Area

The “Visual Results” section, located in the lower left of the interface, presents the processed version of the analyzed architectural plan. In this area, the system renders detected elements directly onto the plan. Each candidate object is highlighted with colored lines, boxes, labels, or ID numbers. The primary function of this section is to facilitate the user’s visual inspection of the system output. For instance, if a window is misidentified as a door, the user can detect this error on the plan. Similarly, if a wall segment is incorrectly marked as an axis point, such discrepancies become readily apparent in the visual results display.

3. Detection List

Adjacent to the visual results, the “Detection List” section presents all candidate elements detected by the AI in a tabular format. This transparency allows the user to understand not only what the system has found but also the confidence level associated with each detection. This list is vital for manual adjustments. The user can select and remove low-confidence candidates, correct misclassified elements, or restart the analysis using only the verified candidates.

4. Confidence Score Logic

A critical aspect of the system is the confidence score, expressed as a percentage, which indicates the AI’s level of certainty regarding each finding. Examples include: • 96% confidence score: High certainty in the detection. • 75% confidence score: Likely correct but warrants verification. • 45% confidence score: Questionable detection that may require removal or further investigation.

5. Removing Candidates Below the Threshold

In the upper left area of the interface, the “Remove Candidates Below the Score” function enables automated data refinement. The user specifies a percentage threshold, and the system eliminates all candidates with confidence scores below this value from the results list. For example, if the threshold is set to 50: • Candidates scoring below 50% are removed. • Candidates scoring 50% and above are retained. If the threshold is set to 70: • A greater number of candidates below 70% are removed. • The results become more selective and refined. This feature is particularly advantageous in analyses with numerous false positives, as it allows the user to conduct bulk removal without manually deleting each candidate.

6. Finding the Right Result Through Iteration

An important characteristic of the system is the user’s ability to refine results through iterative adjustments. The user is not restricted to a single analysis outcome but can experiment with different threshold values, remove low-scoring candidates, retain selected elements manually, and rerun the analysis with the remaining detections. A typical iterative workflow may include: 1. Initial system analysis is completed. 2. User reviews the results. 3. If excessive false detections are present, the user sets the threshold to 60%. 4. The system removes candidates below 60%. 5. User reassesses the visual output. 6. If necessary, manually deletes additional candidates. 7. User restarts analysis using "Run analysis with remaining elements". 8. A more accurate and refined result is achieved. 9. Final output is generated once the result meets satisfaction criteria. This methodology transforms the system from a fully automated tool into a controlled, interactive, human-assisted analysis platform.

7. User Intervention and Manual Cleaning

The structure illustrated in the diagram indicates that user intervention is an integral component of the system. Users are not required to accept all candidates generated by the analysis. Users should have the capability to: • Delete incorrect detections. • Remove low-scoring candidates in bulk. • Select from a list. • Select objects from visuals. • Remove selected candidates from the plan. • Re-analyze with the remaining correct candidates. This functionality holds particular significance in architectural plans, as some drawings may closely resemble each other. For instance, a thin wall line may be mistaken for an axis line, a door frame may be confused with a furniture line, and a window line may appear as a wall section in certain plans. Therefore, achieving a completely reliable result without user control is challenging.

8. Run Analysis with Remaining Elements

A key feature within the Detection list area is the “Run analysis with remaining detections” button. This function enables a new analysis to be conducted solely with the candidates remaining after the user has performed cleaning. This process is essential because: • Incorrect candidates are excluded from the analysis. • The system processes more accurate data. • The result engine utilizes the refined data. • The final output becomes more reliable. This approach reflects a human-in-the-loop methodology within artificial intelligence systems. The system does not rely exclusively on machine decision-making but is enhanced through human oversight.

9. Conclusion Phase

The “Conclusion” block, centrally positioned in the diagram, processes the cleaned and verified findings to produce the final result. During the Conclusion phase, the system can: • Evaluate the remaining candidates. • Interpret the overall structure of the plan. • Organize architectural elements into meaningful layers. • Prepare necessary data for output production. • Generate the final file structure for DXF, PDF, and PNG outputs. It is important to note that the final result is not solely derived from the initial AI analysis. The outcome is shaped through cleaning and corrections performed by the user, thereby incorporating both AI predictions and user control.

10. DXF/PDF Download

Following the conclusion phase, users can download professional outputs. The “Download DXF/PDF” block in the diagram represents this step. At this stage, the system produces the following files: • PDF: Report and visual result output. • DXF: Technical drawing output compatible with CAD software. • PNG: Processed visual plan output. The DXF output is particularly valuable for architects and engineers, as these files can be opened in AutoCAD, DraftSight, LibreCAD, or similar CAD applications. Consequently, plans processed via artificial intelligence can be seamlessly transferred to technical drawing environments.

11. Generated Files Panel

The “Generated Files” panel, located on the right, enumerates all files created by the system. This panel enables users to open or download result files individually. The panel may include the following file types: • Column and Shear Wall PNG • Column PNG • Shear Wall PNG • PDF • Column and Shear Wall DXF • Candidate Column DXF • Candidate Shear Wall DXF This listing demonstrates that the system produces multiple outputs, each serving a distinct purpose. Column and Shear Wall PNG provides a combined visual representation of all findings. Column PNG visually isolates column candidates. Shear Wall PNG archives the final display presented to the user. PDF files are suited for reporting and distribution. Full Plan DXF offers a comprehensive plan output viewable within a CAD environment. Candidate Column DXF is a specialized technical output containing only column candidates. Candidate Shear Wall DXF is a technical output comprising shear wall candidates.

This explanation is displayed under the third flow diagram according to the selected site language.