Great work. Just Yesterday I thought about LLMzip and asked myself if this is something which could vastly improve HTML compression when done at Google scale and shipped with browsers. I haven't done any research though.
I don't use a extra full-text search for my browser history. Why? Because usually searching the headlines and/or urls is enough.
In order to fulltext-search all the webpages one has to crawl and index all the pages first. While crawling might be easy (since you visit the pages anyways), indexing and then searching is a bit more tricky.
One idea for a useful homebrew plugin might be to make use of Google Custom Search. Since Google is quite good at crawling, indexing and searching this might be something to look into.
I am interested if you come up with something useable.
spell correction is somewhat of the most popular and easy to understand problems where machine learning can help. For example: several user enter a misspelled word and then correct them in the next step. If this happens often enough you might infer that the second word is most probably the correct version, especially if the second word has a small levenshtein -distance to the first.
You can use it this way. But what does the data tell? Group A is definitely performing better than group B or better it is not a random coincidence that group A is better performing.