Custom models are built for a specific domain. Let's take an example:
The general image recognition model can still recognize the circuit board, but it lacks domain specific knowledge. Therefore, important details might be missing.
As another example, from Amazon Machine Learning guides:
The key question is how one can build a domain specific model. First, there are already existing models, for instance, in Amazon ML marketplace, from different vendors like LexisNexis for legal ontology in text analysis. But in general, for something specific, you might want to build your own model.
The tricky part is having enough data because any model you want to build, either recognize your product codes in text or detect your hardware inventory via image, you will need a lot of data. This is precisely where Bitskout can help as using workflows and reward create a mechanism to ensure that any job that happened is captured. Which, in turn, gives you the data you need.