Computer Vision Based Mobile Application For Flaw Detection

In this case study, we will look at a computer vision rapid prototype that was built to scale to thousands of simultaneous users. The product was ideated by our client, a leading global fortune 500. We helped them realize a fully scalable API-backend that took advantage of customized next-gen A.I. Models to deliver anomaly detection via mobile application. The Minimum Viable Product successfully automated and delivered an instant remote service guarantee for a process that took weeks when carried out manually.

investigation report

A private database was provided by the client for remote access. A specific subset of the dataset was chosen to be used for external validation as well as blackbox testing. The dataset was insufficient for high-accuracy results but was evaluated to serve sufficiently as a proof of concept. Work began immediately to create an MVP or Minimum Viable Product to demonstrate the viability of the solution. 

development cycle

The development cycle lasted a total of 10 weeks and 12 sprints. Multiple teams were deployed to enable parallel developments of features/modules. AI models with greater 99% accuracy in flaw detection were developed as the first phase to enabling highly accurate flaw-type classification. Flaw-type classifiers were also built and demonstrated the feasibility of classification on datasets of less than hundred images.

A mobile application capable of detecting anomalies through remote client-supplied images was deployed to validate the concept of automating a previously manual contractor/inspector-driven process. The project successfully demonstrated the business value in the application and was subsequently pushed for full-scale development.

deployment

The models were built into a scalable API that could be deployed on the organization’s private cloud hosted in AWS. A full-scale frontend application was integrated into the application and built to the client’s requirements that satisfied UX requirements. The automated process previously took upto eight weeks could now be performed, recorded, and given initial feedback in a matter of seconds.

 

Contact Us

Thank you for trying to contact KayaDev AI. Please fill out this form and one of our executives will get in touch with you shortly. 

Just so we can contact you back.
Give us your linked and we'll expand your network in a refreshing way!
Tell us whats on your mind and we'll have the right person with the right plan of action reach out to you.

Contact us

Thanks for your interest! Please fill the information below to help us serve you better.