Loss of context midway through processing.

I hope this message finds you well.

I’m experiencing an issue with the AI system where it inconsistently processes content extracted from a PDF or structured source. Initially, it provides well-formatted and accurate bullet points, but after a few entries, the output becomes fragmented and cluttered with metadata (e.g., repeated "Source:" tags, internal file IDs like 7f5fdb5db87b_probava_metabolizmi..., and broken sentences).

In some cases, the text cuts off mid-sentence or includes non-human-readable formatting like “%%Source:” or incomplete quotes. This makes the content difficult to use or interpret.