Case Study · Manufacturing

Where artificial intelligence meets industrial reality

Designing a scalable system for AI-driven weld inspection.

Client
North American machine vision solutions provider
Industry
Industrial automation and manufacturing
Environment
Distributed development team, AI-driven inspection workflows, high-volume image data processing

Objectives

  • Frontier technology. A scalable system capable of supporting high-volume image data, model training, and deployment within a unified environment.
  • Operational efficiency. A way to reduce time spent managing and moving data so engineering effort could shift toward AI model development and refinement.
  • Long-term ownership. A solution the internal team could understand, maintain, and expand as new AI use cases emerged.

Outcomes

  • Hundreds of thousands of images handled within a unified system for training and analysis
  • Months of manual effort reduced by eliminating fragmented data handling and coordination
  • 5–6 months from kickoff to a functional, deployable AI platform
  • 4+ model types supported including detection, segmentation, classification, and anomaly detection

The service-oriented architecture we implemented wasn't just the right choice for the problem at hand: the customer is now modernizing other components onto the same infrastructure.

— Aaron Mavrinac Co-Founder, Opreto
01

The Opportunity

One of North America's largest machine vision providers saw an opportunity to make their weld inspection process faster and more reliable with AI.

The client's internal team brought deep domain expertise but limited experience with distributed, containerized systems. The need for a more structured and scalable approach to data and model workflows became clear.

Opreto partnered with them to design and build a platform that brought structure to data workflows and simplified model training. Over time, the team moved from disconnected processes to a more unified system that supports faster iteration and ongoing improvement.

02

The Challenge

The problem? Training the AI involved manually moving hundreds of thousands of images through a fragmented pipeline. Images had to be matched to production outcomes, tracked in spreadsheets and folders, and moved between systems before they could be labeled and used for training.

Data lived in multiple places, and the entire process took time, introduced risk, and pulled engineering focus away from improving models.

03

The Solution

Opreto supported the work through several key initiatives.

Designing architecture to support required data workflows

Opreto helped design a unified system to manage high volumes of image data and make it accessible across teams.

  • Images routed from production into a server-based environment
  • Browser-based access for labeling and review
  • No file transfers or local storage dependencies
  • Reduced versioning issues and manual coordination

"Everything lives and persists on the server, and it's much easier to manage."

Director of Engineering

Establishing a scalable, containerized platform

Opreto set up a service-based architecture designed to scale with the team's AI workflows. Using Docker, APIs, and built-in monitoring tools, the platform connected data, training, and system operations in one place. It also introduced a new way of working for the internal team and created a foundation they could build on as new use cases emerge.

Connecting labeling, training, and deployment workflows

Opreto brought together previously disconnected steps into a single, continuous workflow:

  • Access to images within the workflow
  • One-click trigger for model training workflows
  • Automated movement of outputs into a versioned model registry
  • Ability to generate predictions and iterate on models quickly
  • AI-assisted labeling using model predictions to accelerate annotation

"I labeled the images, hit train, and everything ended up in the right place. All the services were connected."

Director of Engineering

Developing collaboratively and enabling long-term ownership

Opreto worked closely with the internal team throughout the project, with shared visibility into progress and regular input from developers. Standups and reviews kept everyone aligned and helped catch issues early as the platform evolved. By the end, the team had the knowledge and confidence to maintain and extend the system on their own.

04

A smarter foundation for AI, and a stronger future ahead

The platform is now in use and continues to evolve. What started as a way to support weld inspection has grown into a foundation for a wider range of AI-driven inspection workflows.

The biggest impact? The platform replaced a patchwork of tools and manual steps with a single, consistent workflow. Data, labeling, training, and deployment now live in one system, reducing manual coordination and making it easier to iterate on models. Engineers can spend more time improving performance and less time managing data.

The benefits go beyond a single use case

  • One place to manage high-volume image data
  • Faster labeling and training cycles
  • Support for multiple model types and combined AI approaches
  • A platform that can be used across future customer deployments
  • A foundation for applying AI to new inspection use cases

"All the services were talking to each other, and everything ended up in the right place — that's when it really clicked."

Director of Engineering

Explore More

Computer Vision & Machine Learning

The detection, segmentation, classification, and anomaly-detection expertise behind production AI inspection systems.

Manufacturing Software Development

Scalable, containerized systems that bring AI and automation to the factory floor.

Building AI-forward infrastructure?

Let's talk about turning fragmented data and model workflows into one scalable platform.

Let's Talk