Recently I started experimenting with Swift and deep learning using the Swift for Tensorflow toolchain. It seems that the support for jupyter notebooks can be improved. https://github.com/google/swift-jupyter I tried to use this kernel with the XCode toolchain on MacOS which did not work out of the box. So I used the provided Dockerfile to build a container. It works but I couldn't connect to the instance in VSCode and executing basic operations seems too slow. Having type information in jupyter would be helpful as well. When executing the notebooks in VSCode this should also be possible. This is the main idea I propose to work on during GSoC.
Besides this I could create new and improve existing libraries like SwiftPlot (e.g. interactivity, more chart types).
Another area I'm very interested in is cross-platform ui programming and 3d stuff. I think it would be great if we could use swifts emerging deep learning features in end user programs to improve user experience. https://github.com/Cosmo/OpenSwiftUI There is some work being done in that direction already, I could help with this.
Although these are not in the project ideas list, is someone willing to mentor me on any of these ideas?