Agentic optical inspection for complex hardware
Spotting scratches or bent pins on complex hardware like printed circuit boards is easy. The hard part is multi-step root cause analysis when failure modes show up. This requires engineers to validate against computer-aided design (CAD) files, run multiple image analytics models, and reconfigure workflows on the fly. Most inspection tools handle one piece of that problem. Few handle all of it in a natural language prompt.
This demo shows how an agentic optical inspection solution closes that gap. It connects to CAD systems, inspection knowledge databases, and anomaly detection models, then lets engineers describe an inspection workflow in plain language. Reconfiguring a process for a new board or a new defect type is now a conversation instead of a redevelopment cycle.
Watch how it works on a real scenario: a batch of malfunctioning printed circuit boards (PCBs) that need manual root cause analysis. The walkthrough covers designing the board in a CAD tool, receiving images of the failed units, and configuring the agentic application with three tools: one for reading CAD data, one for image processing, and one for anomaly detection.
Two skills do the heavy lifting from there. An image stitching skill combines multi-camera captures into a single board image, built entirely from a natural language prompt. An anomaly detection skill goes further: it overlays the CAD schematic on the input image, registers component positions, crops each one, and runs anomaly detection automatically. An operator can review hundreds of components by running a single skill instead of hundreds of manual checks.
The same approach extends to knowledge base search, image processing, CAD data, and multi-step reasoning, combined into workflows that support both manual troubleshooting and fully automated production lines.
See how agentic optical inspection turns a multi-tool, multi-step inspection process into a single natural language command.
