AI Scaling with TwinKnowledge and AWS Collaboration
Introduction
The architectural, engineering, and construction (AEC) industry is undergoing a digital revolution, with artificial intelligence (AI) and machine learning (ML) at the forefront. TwinKnowledge, an innovative AEC AI company, is redefining how construction drawings are designed and reviewed at scale.
In collaboration with Amazon Web Services (AWS) Prototyping and Cloud Engineering (PACE) team, TwinKnowledge addressed a crucial challenge: scaling their AI-driven computer vision (CV) platform to process thousands of architectural drawings efficiently. This solution leveraged model fine-tuning with AWS services, enhancing large language models (LLMs) and CV models to improve construction document accuracy.
TwinKnowledge’s mission is to minimize errors in construction drawings and boost project efficiency. However, they faced a scalability limitation that restricted their ability to implement a broader AI strategy. With Amazon SageMaker’s MLOps capabilities, AWS helped TwinKnowledge scale model fine-tuning across clients while maintaining high performance and accuracy.
This blog explores how this solution not only solved scalability constraints but also laid the foundation for enhanced AI-driven insights in construction document processing.
1. The Industry’s Information Problem
Challenges in Construction Drawings
Two-dimensional (2D) drawing sets remain the primary reference in construction projects. Over 70% of AEC professionals still use printed blueprints, leading to inefficiencies and errors. To mitigate these risks, companies invest in quality assurance (QA) and quality control (QC) teams, yet mistakes persist due to the complexity of these documents.
A PlanGrid-FMI industry report highlights that nearly 25% of construction rework stems from incorrect project information, costing the U.S. industry $14.3 billion annually. Professionals spend:
5.5 hours weekly searching for project data
5 hours weekly resolving information-related conflicts
4 hours weekly fixing errors and rework
These inefficiencies demand an AI-driven model fine-tuning approach to automate document review and enhance accuracy.
TwinKnowledge’s AI Solution
TwinKnowledge integrates LLMs and CV models to process textual and graphical data in construction drawings. However, off-the-shelf AI models failed to accurately interpret drawing set structures, requiring a customized model fine-tuning approach.
Their proprietary AI pipeline first extracts textual and graphical information separately. Then, a fine-tuned CV model maps design details to lessons learned, requirements, and standards. However, scalability remained a challenge, requiring an automated AI scaling infrastructure to support their expanding client base.
2. Scaling Model Fine-Tuning for AI Workflows
AI Challenges in Construction Compliance
The AEC industry struggles with compliance verification due to:
1. Massive amounts of project data that must be reviewed
2. Limited AI models capable of mapping design compliance effectively
Model Fine-Tuning for AI Scalability
TwinKnowledge leverages model fine-tuning to customize AI models for different clients and projects. Their solution transforms graphical data into structured knowledge and embeds it within LLMs to provide real-time, AI-driven insights.
The architecture includes two distinct AI pipelines:
1. Text Processing Pipeline – Extracts and processes project data from contract documents
2. Computer Vision Pipeline – Processes drawing sets to map design structures
The result is a scalable AI infrastructure, enabling model fine-tuning to achieve near-human accuracy in drawing compliance checks.
AWS-Powered AI Scaling
To support this model fine-tuning approach, TwinKnowledge collaborated with AWS PACE to design an MLOps pipeline with three key components:
1. Data ingestion and labeling – Automates document classification for AI training
2. Training pipeline – Fine-tunes AI models for high-precision document review
3. Inference workflow – Deploys trained models for real-time construction analysis
By implementing this MLOps strategy, TwinKnowledge achieved automated model fine-tuning, enhancing AI-driven workflows across clients.
3. AWS-Powered Infrastructure for Model Fine-Tuning
Data Ingestion and Labeling with AWS
For AI models to deliver accurate results, high-quality, well-organized data is crucial. TwinKnowledge built an AI-driven data pipeline using AWS services to automate document processing. The workflow includes:
1. Construction document upload via AWS API Gateway
2. Authentication with Amazon Cognito
3. Data storage and metadata processing using Amazon S3 and DynamoDB
4. Inference execution with AWS Lambda and Amazon SageMaker
5. Model monitoring using AWS CloudWatch and X-Ray
By automating data ingestion, TwinKnowledge accelerated model fine-tuning while maintaining high AI accuracy.
Fine-Tuning Computer Vision Models
AWS’s scalable MLOps infrastructure allows TwinKnowledge to:
- Label key drawing set elements (title blocks, details, schedules)
- Fine-tune CV models to adapt to project-specific drawing sets
- Deploy inference pipelines for real-time compliance checking
By leveraging AWS Inferentia2-powered SageMaker instances, TwinKnowledge reduced AI model inference costs while improving processing speed.
Conclusion
Through this AWS collaboration, TwinKnowledge successfully scaled model fine-tuning for AI-driven document review in the AEC industry. The solution not only improved accuracy and efficiency but also laid the foundation for future AI advancements in construction compliance.
Key Takeaways
- Model fine-tuning enables scalable AI solutions for complex AEC workflows
- AI-augmented document reviews enhance, not replace, human expertise
- A flexible AI scaling infrastructure supports rapid industry adoption
As the AEC industry continues to digitize and embrace AI, TwinKnowledge’s AI platform is set to drive efficiency and innovation. By integrating MLOps best practices with Amazon SageMaker, companies can scale AI workflows seamlessly and enhance project accuracy.
Cloudastra supports AI initiatives by providing relevant cloud services, ensuring organizations leverage cutting-edge AI scaling infrastructure to boost productivity and streamline processes.
Do you like to read more educational content? Read our blogs at Cloudastra Technologies or contact us for business enquiry at Cloudastra Contact Us.