Executive Summary
A US construction tech startup partnered with Elastiq and Google Cloud to deploy Elastiq Pixels, a semantic image search solution enabling analysis of construction site images across millions of photos to ensure employee safety compliance-reducing search time from hours to seconds.
Industry Context
Construction has the highest number of deaths and injuries compared to other US industries. Yet the sector has been “among the slowest to adopt new technology.” This gap between safety importance and technological capability represents both a challenge and an opportunity.
Client Profile
A leading construction-tech startup that empowers construction teams through rewards and recognition programs. The platform integrates with major construction management software and uses AI to help companies optimize performance and improve safety outcomes.
Business Problem
Three core challenges prevented effective safety monitoring:
Manual Review Bottleneck
Traditional image analysis required manual review of every photo, creating time-consuming, error-prone processes that delayed safety assessments and reward distribution.
Keyword Search Limitations
Keyword-based searches failed to accurately identify safety violations. Low accuracy and high false positives required significant manual intervention.
Lack of Multi-Modal Search
Existing systems couldn’t effectively search across image content, metadata (dates, locations), and user information simultaneously, limiting comprehensive safety analysis.
Elastiq Solution: Elastiq Pixels
We deployed a comprehensive semantic image search platform:
Semantic Search with Vertex AI
Leverages Google Vertex AI for multi-modal embeddings enabling semantic search across millions of images. The system identifies safety violations beyond simple keyword matching-understanding context, equipment, and behavior.
Multi-Modal, Multi-Dimensional Search
Proprietary search logic enables queries across:
- Image content (what’s visible)
- Metadata (when, where, who)
- Project information (site, phase, contractor)
Scalable Architecture
Designed to handle billions of images with sustained high performance using:
- Google Kubernetes Engine (GKE) for compute
- OpenSearch Vector Database for fast similarity search
- Cloud CDN for global image delivery
Results
The impact was immediate and measurable:
- Search time reduced from hours to 1-2 seconds
- Automated compliance checks eliminated manual photo tagging, saving 100+ person-hours monthly
- Real-time image verification enabled 10x faster reward distribution for safety compliance
How It Works
Safety Query Examples
The system understands natural language safety queries:
- “Workers without hard hats near heavy equipment”
- “Missing safety barriers on elevated platforms”
- “PPE compliance in welding areas”
Each query returns relevant images ranked by confidence, with highlighted areas of concern.
Continuous Learning
The system improves over time as safety managers validate results, building a feedback loop that increases accuracy for site-specific safety patterns.
Conclusion
By combining computer vision with semantic search, Elastiq Pixels transforms construction site safety from reactive to proactive. Safety managers can now identify compliance issues across millions of images in seconds, enabling intervention before incidents occur-not just documentation after the fact.