Hello Shipla,

I would suggest to ask LLM to output a json with three keys.

1. Highlights

2. Type of issue or issues

3. Flag for the type of issue you want to filter.

This would enable you to have an json with all the necessary outcome.

Hope this helps :)