Executive Summary
A leading US-based construction tech startup faced a critical challenge: How do you ensure safety compliance of employees at a construction site across millions of photos?
They partnered with Elastiq and Google Cloud for our semantic image search solution, Elastiq Pixels. This allowed them to empower their clients to analyze the construction site images. Thus, transforming how the industry approaches safety.
Client Profile
Our client is a leading construction-tech startup that empowers construction teams to change employee behaviors through the use of rewards and recognition. Their platform integrates with leading construction management softwares and they use AI to help construction companies optimize performance and improve safety outcomes.
Business Problem
Year after year, the statistics collected by the U.S. Bureau of Labor Statistics demonstrate that construction is riddled with risks to safety, with the highest number of deaths and injuries compared to other industries in the U.S. This industry has also been among the slowest to adopt new technology as highlighted in a recent report by McKinsey research.
Our client wanted to work on this with their safety rewards program. However, they faced significant challenges in analyzing their vast image repository:
- Manual Review Bottleneck: Traditional methods required manual review of every image, a time-consuming and error-prone process, hindering timely safety assessments and reward distributions.
- Limitations of Keyword-Based Search: Keyword-based searches proved inadequate for accurately identifying safety violations within images. This resulted in low search accuracy, high false positive rates, and significant manual intervention, increasing operational costs.
- Lack of Multi-Modal Search Capabilities: Existing systems lacked the ability to effectively search across multiple dimensions, such as image content, metadata (date, location, etc.), and user information. This limited the scope of safety analysis and hindered real-time insights.
Elastiq Solution
Elastiq created a special system called “Elastiq Pixels” that deployed on Google Cloud. Here are some of the important things it could do:
- Semantic Search with Vertex AI: Elastiq Pixels leverages Google Vertex AI to generate multi-modal embeddings, enabling semantic search across millions of construction site images. This allows for identification of safety violations beyond simple keyword matching.
- Multi-modal, Multi-dimensional Search: Elastiq Pixels incorporates a proprietary search logic that enables powerful multi-modal search capabilities. This allows users to search across various dimensions, including image content, metadata (dates, locations, users), and project information, providing comprehensive insights into safety compliance.
- Scalable Architecture: The Elastiq Pixels platform is designed to scale seamlessly to handle billions of images as the client’s image repository grows. This ensures continued high performance and rapid search results even with massive datasets.
The Elastiq Pixels platform utilizes a comprehensive Google Cloud technical stack:
- Machine Learning: Google Vertex AI for image multi-modal embedding generation.
- Container Orchestration: Google Kubernetes Engine (GKE) for automated scaling and management of the entire platform including Opensearch Vector Database
- Cloud Storage: Google Cloud Storage (GCS) buckets for secure image file storage
- Content Delivery Network: Google Cloud CDN for fast and reliable global content delivery
- API Gateway and Load Balancing: Google Cloud Load Balancer for streamlined API access and workload distribution
Google Cloud’s comprehensive suite of services provided the foundation for a reliable, scalable, and cost-effective solution, enabling Elastiq Pixels to deliver real-time insights and improve safety outcomes for construction companies worldwide.
Results
Elastiq Pixels platform helped the customer enhance image management.
- Reduced image search time from hours to 1-2 seconds
- Automated Compliance checks and eliminated manual tagging of photos saving 100+ person-hours of manual work each month.
- Real-time verification of images enabled 10X faster reward distribution
This partnership shows how semantic search capabilities can enhance compliance verification.