Show HN: AI Text Rewriting | Hacker News Reader

Show HN: AI Text Rewriting

uglot.ai

8 pointsmanx8 comments

Does it work one sentence at a time or can also do a paragraph? Looks like it works one sentence at a time. It ignores next sentence onward. Also, punctuation seem to be weak, it didn’t correct questions that I ended with period (.).

Can it be easily extended to take a text sample and rewrite a different text sample using the keywords from first text sample? Most probably not possible.

Use case: any situation where incorrect statement might be manually or automatically written.

Potential users: language learners, email writers, messaging app users, people using translation app to translate then use your service to make translated statement better, people using voice to text/dictation, correcting OCRs, ...

Developer here. When writing, I was not satisfied with existing grammar checkers (Grammarly, LanguageTool). What I really wanted was better wording for what I wanted to say. My NLP experiments turned into a working product that I now use myself. I am looking for ways to easily monetize. Currently, I'm thinking about a prepaid solution instead of a monthly subscription. You would pay per character until your paid amount is used up. Any other ideas?

Appreciate any feedback

I like it!

https://i.imgur.com/qAfcgzH.png

Can't comment on the monetization, but I have an idea for it- A lookup function such that the following could be accomplished:

You take a PDF of a real-world Book, that only ever existed in physical book-format prior.

Have your system read the book, and rewrite it with links and information on the content of the book. For example, it could search proper nouns for people places things that have a wikipedia page, and makes those instances a link to their respective wiki.

Additionally, events of consequence could also have links to their wikis for drill down on things spoken of in the book.

Another interesting aspect would be for it to build a list of characters in each book read, and being able to correlate series based on the recurrence of people places things in common between volumes.

Thank you for your ideas! I think you're overestimating a bit what it is able to do. Or am I missing something? How can these ideas be accomplished using text-rewriting?

a developer API would be great

Works great actually. Did you source bitext for each language, or is it all synthetic?

It's purely synthetic.

It works surprisingly well for synthetic data. Is it a pre-trained transformer in the backend?

I have a competing side-project in the same space, but using bitext and a vanilla transformer.

Reply on news.ycombinator.com