Hi @Linnie,
Sounds like you are having similar trouble to this user. I think the funny quality plot that you are seeing is technically okay — it looks like maybe what is happening is the read joining method used by the company is assigning an arbitrarily high PHRED score to the overlapping bases. The non-overlapping bases are unadjusted, leaving you with a goofy looking plot. But I think that is okay! (though the lack of real quality scores may impact the ability of methods like dada2/deblur to detect noisy sequences within this overlapping region 
)
Check out this tutorial on working with joined paired-end sequences — note that the quality plots on that page look similar to yours.
I hope that helps! Good luck!