Hybrid architecture that combines cloud training and the edge enables on-the-fly learning.
Elementary, a machine vision inspection company based in Los Angeles, has unveiled its new VisionStream, an AI inspection system that learns directly from production lines.
The company says it’s new technology flags defects within seconds and requires no labeled data, vision expertise, or production downtime.
“Traditional inspection systems require you to shut down production during setup, but factories can’t afford that downtime,” said Arye Barnehama, CEO of Elementary. “VisionStream overcomes this challenge by learning while production runs.”
Citing Rockwell’s 2025 Smart Manufacturing Report, which indicates many Artificial intelligence and machine learning (AI/ML) pilots stall in the data management phase, elementary says its new product removes a key obstacle to scaling AI in manufacturing: the operational burden. First-generation AI turns factories into data managers, requiring weeks or months of data collection, review, and tuning, diverting resources and scarce engineering talent from other critical areas.
VisionStream eliminates this by learning through observation. In lab testing, it watched parts move down the line and reaches up to 99.9% accuracy within seconds. It requires no data preparation or line stoppage.
In one test, VisionStream took just 12 seconds to learn what “normal” spark plugs look like and then immediately flagged an electrode defect missed by human experts. The company
The technology was developed with five key capabilities:
- Live Learning – Captures and learns from real production data without staged defects or operator input
- Edge Processing – Runs locally for real-time results while syncing securely to the cloud
- High Accuracy – Detects up to 99.9% of defects, including subtle or unexpected flaws
- Operator Oversight – Incorporates human feedback to improve performance over time
- Universal Integration – Installs with new or existing cameras and connects to PLC, SCADA, MES, ERP, and BI systems
This rapid learning capability enables inspections that were previously impractical, such as complex or rare defects, high-mix production, short run production and urgent quality issues when you don’t have time to implement a brand new quality system.
This deployment speed is made possible by its hybrid architecture. Foundational inspection models train in the cloud, while edge neural networks adapt to each production line. This approach combines broad defect knowledge with on-the-fly learning from live production data.
