NativeJars bundles those libraries so that they can be easily imported into Java projects. Sure, you could take those same libraries, zip them up and put them in a JAR file yourself. Or you could just put it in Docker. But sometimes you need native libraries in a JAR file, and NativeJARs makes that easy.
Once those libraries are packaged in the JAR, NativeJARs lets you call native functions from Java. For example, let’s say there’s an evaluation function for machine learning that someone wrote in TensorFlow. Well, you don’t have to reimplement it now, you can just call it. Need a beam-search implementation that’s only available in C++? Just call it.