nadavweisler / communitale Goto Github PK

The service get paragraph and image and create book page

Extract the base seed messages from the code into a json, clean up the seed from unnecessary messages and sort the messages for easier tuning and debugging.

The way we split the title from the story causes an exception due to inconsistent formatting from GPT's response

Need to rephrase GPT's prompts and parsing methods to address this issue.

The function need to get a prompt and return paragraph and image for each page

Delete or put in directory that in .gitignore

Currently the backend saves a requested book's assets as files and then re-reads them to generate the final output pdf. The pdf is again stored on the backend as a file and passes the file's location to the front end for download. This creates a few problems:

1. The server can only generate one book at a time (two people entering the website will cause issues)

2. Adds unnecessary complexity to the backend.

Proposed solution: keep image data in memory without saving the png files to the server. And add unique identifiers for each request that will be used in the pdf output's name. This way multiple pdfs can be generated at the same time, returned to the end-user correctly and removed safely at the end of the transaction without conflicting with other requests.