CS491 11/7
Meeting Overview
Participants
Team members and participants briefly greet each other and express their well-being.
Hard Harjaval is unable to join the meeting but may join later.
Agenda Overview
Overview notes are to be shared in the chat for reference, with an option for screen sharing available.
Focus on multisource data correlation and JSON to IAC entity conversion.
Multisource Data Correlation
Overview
The team discusses a system for correlating data from multiple sources using a confidence scoring mechanism.
A graphical representation showing the workflow is mentioned, which includes:
Multi-source data inputs (LLMs, models)
Data correlation engine
Conflict detection
Entity resolution
Confidence scoring module
Outputs into a master data record
Final integration into a confidence annotated output pathway leading to audit trails and human review.
Conflict Definition
A conflict arises when multiple inputs present different values for the same data point.
The resolution would depend on confidence scoring or reference to the most utilized or credible source.
Confidence Scoring Explained
The confidence score is calculated using various weights:
Source Reliability: Higher reliability from complex schematics (e.g., Building Information Modeling - BIM) receives more weight compared to simple floor plans or images.
Data Quality: Higher quality or more complete data structures (e.g., JSON schema) are weighted higher.
Consistency: Data must align without contradictions (e.g., measurement errors).
Precision: Greater precision results in a higher confidence score.
Corroboration: Agreement across multiple sources increases weight.
Statistical Validity: Ensure data adheres to logical and building code standards.
Semantic Coherence: Overall, the data should make sense logically.
Further refinement of the scoring module is ongoing.
Merging Results
The team explores merging results using either:
Additional orchestration services or a specific LLM to execute the merging process.
Preference for using an LLM for merging due to ease, with orchestration as a backup in case of inaccuracies.
Need for testing different instances or variants of the same LLM to verify independence of the data during merging.
Feedback and Workflow Validation
Feedback on the proposed workflow appears positive with ongoing testing and refinements acknowledged.
The team should examine potential roadblocks during further research.
JSON to IAC Entity Conversion
Approach Outline
There are two proposed approaches for converting JSON data into IAC entities:
Direct Scripting Approach
Utilize a Python script or orchestration service (such as IFC OpenShell) to convert custom JSON into an IFC file.
Pros: More tailored to project needs, builds in audit tracking, and allows precision for specific schematic inputs like plumbing, electrical systems, etc.
Cons: Scripting complexity may lead to issues with changes in JSON schema requiring script adjustments.
Backup Approach
Employ a CityJSON converter, which handles more structured input to produce an IFC file.
Pros: Easier implementation, existing documented converters.
Cons: Less precise for individual buildings as CityJSON is designed for broader cityscapes.
Further testing will elucidate which method is most effective.
Future Steps and Testing
Upcoming Actions
Establish validation mechanisms in the IAC XPOS to see how well it integrates into three-dimensional modeling software.
Demonstration of sample IFC files from Revit models for comparison and understanding of required structures.
Consider adding additional description text as an improvement to image-to-3D translation tasks.
Research current standards for wall thickness, door, and HVAC specifications to enrich data input quality.
General Updates and Communication
Team Communication
A satisfaction report will be requested from team members to evaluate performance across technical, scheduling, and cost management dimensions.
Feedback should be sent directly to Professor Larry Bailey.
Meeting Schedule
Next meeting is proposed for two weeks from now on a Friday, coinciding with the lead-up to Thanksgiving. Confirmed access to necessary cloud environments for project advancement.
Conclusion
Acknowledgment of team efforts and encouragement to keep progressing on ongoing and future tasks. Key reflection on real-world problem solving within the project context.
End of meeting with positive affirmations and plans for follow-up communication.