Vision inspection and the performance of AI truly accessible to everybody

AI-go can be used to solve a wide range of quality control tasks, like classification, segmentation and OCR (optical character recognition), achieving higher performance on use cases covered by traditional technologies and reducing the set-up effort from weeks to minutes.

AI-go allows the user to autonomously and easily create vision inspection models in a few minutes and without the need of resorting to external specialized integrators, or without previous knowledge on AI and other traditional vision inspection techniques.
Differently from other AI-systems, models can be trained using only a few examples (10 - 40), reducing the set-up effort and enabling a highly replicable approach characterized by a short time-to-value.

The AI-go suite consists of two components:


A cloud platform with an easy-to-use graphical interface that provides a simple model creation for the following tasks:

  • classification: both binary (good vs bad) and multi-class (defect 1 vs defect 2 vs defect 3) CASE STUDY
  • segmentation (highlighting defects and their shape in an image) CASE STUDY
  • OCR (optical character recognition) CASE STUDY


An edge component (installed on the production line and which can also work without internet connectivity) for high-performance model serving. This is where the AI models can be installed and the prediction easily generated.

AI-go STUDIO – Model management platform

  1. Collect images from the production line: 10-40 images per class.
  2. Upload images and label images: add tag to images (e.g good / bad or defect1 / defect2 / defect3…).
  3. Training model autonomously: just press one button!

AI-go STUDIO – Model management platform 

  1. Test and validation: check the robustness of your model on new set of images and make sure of the performance before going to the production line.
  2. Download trained and validated model from AI-go Studio.


  1. Load trained and validated model from AI-go Studio to AI-go Runtime and deploy on edge device.
  2. Inference: process real time incoming data to predict the image result.
  3. Production statistics: real time extended diagnostic.



Designed to ensure the success of your AI applications throughout their entire life cycle.

  1. Monitoring: continuously extract information about the performance of your models in production, get dashboarding and alerting at any level and set up your remote «control room».
  2. Governance / compliance: expands the monitoring through a constant advice service from Orobix experts. Guided retraining and scheduled reporting to keep the performance of your production at the maximum level.
  3. On-demand: with direct contact to Orobix you can access on demand to:
    • custom backbones
    • model fine-tuning
    • test and validation

The AI-go vision inspection suite allows clients to autonomously develop and deploy their own models, which are able to solve mission-critical challenges and are suited for industrial-grade applications, without having to always rely on external integrators and custom project-by-project approaches.

Furthermore, AI-go suite is also designed to help machine manufacturers to add new functionalities to their machines, being able to offer a new generation of services to their own customers.

AI-go makes possible to:


AI-go allows clients to develop on their own AI models for a wide variety of computer vision problems, learning all the variability directly in production:

  • Classification involves predicting which class an item belongs to, both binary (e.g. good vs bad) and multi-class (e.g. defect1 vs defect2 vs defect3). AI-go supports both full-image classification models (CASE STUDY) and in patch-based classification models (CASE STUDY).
    Classification is used to solve problems like defect identification, anomaly detection, presence / absence, object detection, assembly verification, etc.
  • Segmentation involves dividing images into segments representing objects, their parts, or the shape of a defect (e.g. to get its size). Image segmentation is used to defect sorting / qualification, shape analysis, etc. CASE STUDY
  • OCR is the recognition of printed characters even in case of curved and uneven surfaces (e.g. vials, flasks, bottles, jars, bags, blister and tubes) or irregular print quality, or embossed writings (e.g. dotted and industrial fonts) CASE STUDY


Automatically select the most suitable pre-trained model for solving the specific problem, train it on a few examples (about 10-40 images) and put it into production in complete safety, dramatically reducing the set-up effort (time, money and skills needed).
More intuitive configuration and setup, thanks also to the user friendly interface, less experience and time is needed to achieve good performance.
Run models even on devices with reduced computing power (IoT, edge computing).


Have a solution designed to operate in industrial contexts, considering the aspects left out by generalist solutions (e.g. cycle times, inference speed, reliability, unstable connectivity in production sites, employee training on the line) and easily integrated into pre-existing vision systems.
Remotely check the models to ensure they are working correctly and start automatic re-training to improve their performance over time.
Manage the validation of new models before putting them into production, to ensure that the performance is always in line with the business expectations.

All in one smart camera


Co-developed in partnership with IMAGO Technologies. Processing, optical acquisition and lighting systems all integrated in a single device. Designed for less demanding and slower use cases, VisionCam AI-go is also compatible with AI-go Studio to address more complex problems.


Entry level (1):

  • presence / absence
  • part integrity
  • assembly verification
  • part location (correct positioning)
  • products type classification
  • wrong colour detection
  • macroscopic defect detection



  • directly on the device;
  • compatible with AI-go Studio to leverage more advanced model specializations;
  • less than 1 minute to train a model.


  • entry level models (1) ~ 100 ms

Industrial PC + GPU (optional) and vision system


An industrial PC + GPU (optional), optical acquisition and lighting systems.

The customer can: 

  • autonomously train new models from only few images on AI-go Studio;
  • manage all the models and monitor the system on one interface (AI-go Runtime).
  • solve complex vision inspection problems in the manufacturing industry.
  • achieve better speed and performance.


Entry level (1) and advanced (2):


  • on AI-go Studio;
  • test before deploy in production;
  • all the models organized in a unique place;
  • average time to train and test a model ~ 30 min.


  • entry level models (1) < 50 ms
  • advanced models (2) ~ 500 ms (CPU) 
  • advanced models (2) ~ 50 ms (GPU)