Computer vision model deployment
End-to-end deployment of a CV classification model during an applied AI internship at Avanade.
A manufacturing use case required detecting defective screws using camera footage. The challenge was not just training a model but getting it to run reliably on edge devices in a production environment.
Participated in Avanade's internal applied AI program focused on end-to-end delivery rather than isolated modeling. Trained and validated a classification model provided through Azure ML and deployed it to edge devices using Azure tooling and agents.
Evaluated the model using real footage from the production environment to verify correct behavior outside of static test data. The deployment setup was tested against the actual edge device constraints rather than only on controlled hardware.
A deployed model showing how computer vision can reduce manual quality assurance effort. The internship gave first-hand experience of what deployment constraints actually mean: foreign machines, agents, and operational friction that don't appear in research settings.