There are a lot of tactics you can take to come up with things that are novel (in minor ways). I found that if you write you'll discover new things along the way.
* Write up samples of old ideas in new languages (e.g. X in Scala)
* Write up samples using new libraries (e.g. from the Show HN posts)
* Combine two random things together (e.g. PDF.js + Lunr.js = ...). This works well with new libraries, because people like having their work used.
* Thought experiments can work well (what's a feature of language/tool X that you don't like, and how would you fix it)
* Write up sample code that fills the gaps in some libary's documentation (if you work on a tool and have trouble googling the results, this works well, or if the documentation doesn't answer questions in the way you formulate them)
* Find a CS textbook on a subject you find interesting, find interesting problems but apply them to something you're familiar with. I've had good success with this with an NLP textbook, but apply the exercises to data I have access to so they are novel.
* If you look at language based subreddits, a lot of the smaller ones are starved for content. You can find popular examples from language X and rewrite them for language Y.
* Find a book of puzzles (soduku type things) and write up puzzle solvers.
* Write up samples of old ideas in new languages (e.g. X in Scala)
* Write up samples using new libraries (e.g. from the Show HN posts)
* Combine two random things together (e.g. PDF.js + Lunr.js = ...). This works well with new libraries, because people like having their work used.
* Thought experiments can work well (what's a feature of language/tool X that you don't like, and how would you fix it)
* Write up sample code that fills the gaps in some libary's documentation (if you work on a tool and have trouble googling the results, this works well, or if the documentation doesn't answer questions in the way you formulate them)
* Find a CS textbook on a subject you find interesting, find interesting problems but apply them to something you're familiar with. I've had good success with this with an NLP textbook, but apply the exercises to data I have access to so they are novel.
* If you look at language based subreddits, a lot of the smaller ones are starved for content. You can find popular examples from language X and rewrite them for language Y.
* Find a book of puzzles (soduku type things) and write up puzzle solvers.