I’ve read about microsofts IntelliCode that aims to guide developers with better code completion and I want to start something similar for the Rust programming language as project both to learn ml and to use it at the end. I know that means a lot of work and the result will probably be in a whole different league than what microsoft has build.

​

From what I’ve read syntax checks can be implemented with autoencoders and LSTM networks, but I don’t know how to suggestion for code corrections.

LSTM networks are used for autocompletion and text generation, but I would need to implement a context aware network that gets code from before and behind the cursor. I guess I would need to tokenize the different syntax elements like natural language in NLTK is processed?

Right now this is mostly guess work for me. I would highly appreciate tipps on the network type and design and information about related work.