UnitX CorteX AI Visual Inspection

UnitX is a robotics and machine vision company founded in 2018 by engineers from Stanford, MIT and Google, with an office in Milpitas, California. It automates in-line visual inspection for manufacturers whose defects vary in shape, size, position and appearance, a task that rule-based vision and human inspectors handle poorly.

CorteX is the AI layer of the UnitX stack: users label images, train models with as few as five images per defect type through a drag-and-drop interface, tune thresholds and push models to production lines from a central console. It can run on its own with existing cameras, together with the OptiX software-defined imaging system, or inside turnkey inspection equipment. Target sectors are automotive, EV and lithium batteries, electronics, PCBA, connectors and semiconductors.

• Few-shot training: builds defect models from as few as five images per defect type.
• No-code workflow: labeling, training, threshold tuning and line data are handled through a drag-and-drop interface.
• Pixel-level segmentation: outlines defects precisely and supports classification, counting, measurement and barcode reading.
• Yield simulation: shows expected yield before a model goes live and allows tuning across six quality attributes.
• Central model management: updates and deploys models to lines in several sites without stopping production.
• Line integration: supports major PLC protocols, GigE cameras and an SDK for custom extensions.

An EV battery or cell supplier in northern Vietnam could inspect welds and casings in-line. An electronics or connector plant could detect pin and cosmetic defects on PCBAs. An automotive component supplier could replace manual end-of-line visual checks on machined or cast parts.

Models can be trained from as few as five images per defect type, cutting the data collection needed before go-live. UnitX states that customers see 9x fewer escapes, about 3% less scrap, up to 1.3 million USD returned per line and ROI in under 12 months. A LIMIT decision sets aside borderline parts for review instead of forcing an OK/NG call.

Battery, automotive and electronics supply chains are adding capacity in Southeast Asia, and buyers increasingly require documented, consistent inspection. Few-shot deep learning lets plants automate cosmetic inspection that rule-based vision could not handle, which matters where skilled inspectors are scarce and product changeovers are frequent.

UnitX was founded in 2018 and lists an office in Milpitas, California. It reports more than 190 customers and over 6.1 billion USD of products inspected annually. The website is also published in Japanese, Korean, Vietnamese and Thai. In October 2026 it won first place at the TechGALA Japan 2026 Grand Pitch.