What would you like to see?

I want to upload JSON files and include metadata with each JSON as I am using legal documents. Each JSON {} can be one chunk and the Database can go through a Metadata first to pick correct chunk. The LLM confuses different chunks and mixes them together to give answers. E.g. If I ask what happened in a case - It mixes other case with it and shows issues from there too.
My only solution was to divide the documents into separate txts to get some reliable results although not very perfect as it picks up wrong txt altogether - as it is not pciking up the metadata