checkMate
Carla Cortez�Redwan Hussain�Kam Kasravi�Edward Kirton�Ryan Wilson�
A novel approach to Building Code Compliance with Large Language Models
OUR TEAM
Carla Cortez�Software & Web Dev, Blockchain
Kam Kasravi�Artificial Intelligence, Cloud
Redwan Hussain�Industrial Controls & Automation
Ed Kirton�Data Science, BioInformatics
Ryan Wilson�Analytics, Education Management
David Fenster, AIA LEED AP�Subject Matter Expert, Architecture
MOTIVATION
Address Construction Industry Challenges
MOTIVATION
Digitizing this process can reduce delays and associated overhead costs for firms and prospective homeowners
Target Audience
Architecture, Engineering, and Construction (AEC) firms
VALUE PROPOSITION
Automated Compliance Reports
Building Standards and Codes
Building Information Modeling (BIM) Program Files
�
Compare project files against building codes…
… and auto-generate compliance reports with targeted feedback for reduced errors, faster approvals, and saved costs.
OPPORTUNITY: LARGE LANGUAGE MODELS
MODEL
IFC Schema
Natural Language Text
Building Codes
FUNCTIONAL FLOW
Backend Processing
Intuitive UI
Reports
UI allows easy selection of building elements that can be submitted for code compliance
Building elements are matched with relevant codes. Each element is evaluated for code compliance
A compliance report is returned to the user
APPLICATION WORKFLOW
2
1
checkMate
APPROACH 1: RAG MODEL�Prompt Example
Response
Entry prompt
+
{
"type": "Door Metal, Operable"
"IsExternal": true,
"FireRating": "1.0 hr", "Infiltration": "0.30 m"
}
{ R337.8.3: {
"compliant": True,
"summary": The door is external and it has the correct fire rating
}
}
Document(code='R302.2.1Each townhouse unit shall be separated from other townhouse units by two 1-hour ….
APPROACH 1: RAG MODEL�Architecture Diagram
2
1
APPROACH 1 - MODEL EVALUATION�
Misclassification analysis:
Results:
Accuracy
63%
"compliant": False,
"summary": "The entity does not comply with the building code.
Although the door has a high fire rating of 4.0 hours,
it does not meet the specific fire-resistance rating of not less than
20 minutes when tested according to NFPA 252. The door also does not
meet the performance requirements of Section R337.7.3.1 and SFM Standard 12-7A-1."
Precision
0.78%
Discussion, Pros/Cons of Retrieval-Augmented-Generation
RAG�
PROs
CONs
APPROACH 2: TEXT-TO-SQL
The RAG approach asks a LLM to reason: “Does <wall-1> satisfy <code-X>?”
A LLM can be used to extract information from natural language.
The extracted values may be used to construct SQL queries programmatically.
DOOR
ACC
CRC
ALGORITHM
ZERO-SHOT TEXT-TO-SQL
Section R308 Glazing
R308.6.2: Materials
Glazing materials shall be limited to the following:
SELECT GlobalId FROM IfcWindow
WHERE ( … OR … )
ZERO-SHOT TEXT-TO-SQL
Section R308 Glazing
R308.6.2: Materials
Glazing materials shall be limited to the following:
SELECT GlobalId FROM IfcWindow
WHERE ( IsLaminated = true OR … )
ZERO-SHOT TEXT-TO-SQL
Section R308 Glazing
R308.6.2: Materials
Glazing materials shall be limited to the following:
SELECT GlobalId FROM IfcWindow
WHERE ( IsLaminated = true OR IsTempered = true OR IsWired = true );
AUTOMATIC CODE COMPLIANCE VIA SQL
Passing records:
SELECT FROM <named_entity>
WHERE <defining_attribute>
AND ( <attribute1> <AND|OR> … )
OR ( <exception1> OR … )
Failing records:
SELECT FROM <named_entity>
WHERE <defining_attribute>
AND NOT ( <attribute1> <AND|OR> … )
AND NOT ( <exception1> OR … )
ACC
SQL is very fast!
Text-to-SQL - EVALUATION�
CRC | ENTITY | HUMANS | TEXT-TO-SQL | COMPARISON |
R337.8.3 | IfcDoor | (isExternal == TRUE) AND (FireRating >= 20) AND (isTempered == TRUE OR isLaminated == TRUE OR isCoated == TRUE OR isWired == TRUE) | SELECT GlobalId from IfcDoor WHERE IsExternal = 1 AND (FireRating >= 20) | TTS BETTER |
R302.9.1 | IfcDoor | SurfaceSpreadofFlame<200 | SELECT GlobalId from IfcWallStandardCase WHERE (SurfaceSpreadOfFlame <= 200) | SAME
|
R302.1 | IfcWall | FireRating >= 60 | NONE (tables NYI) | HUMANS BETTER |
R305.1 | IfcWall | NominalHeight >=84 | SELECT GlobalId from IfcWallStandardCase WHERE (NominalHeight >= 84.0) | SAME |
R337.8.2.1 | IfcWindow | FireRating >= 20 | SELECT GlobalId from IfcWindow WHERE IsExternal = 1 AND (FireRating >= 20.0 OR IsTempered = 1 OR SmokeStop = 1) | TTS BETTER |
R308.6.2 | IfcWindow | IsWired == 1 OR IsLaminated == 1OR isTempered == 1 | SELECT GlobalId from IfcWindow WHERE (IsLaminated = 1 OR IsTempered = 1 OR IsWired = 1) | SAME |
DISCUSSION, PRO’S/CON’S OF TEXT-TO-SQL
TEXT-TO-SQL�
PROs
CONs
Useful for driving the co-development of�law and file standards.
LESSONS LEARNED
SAVE TIME AND COSTS
FUTURE-PROOF
Leverage NLP for intuitive prompts and clear feedback responses
Automatically produce detailed compliance reports
Refresh model and constraints as regulations evolve over time
MINIMIZE ERRORS
RECAP: INDUSTRY CHALLENGES�
checkMate VALUE PROPOSITION AND OFFERINGS