I have found using an llm to compress the information instead instead of reading it directly saves time. It's potentially lossy but it's not like the slop you got is going to lose much value by compressing it.