This is really weird for me, a few years ago I was trying to build exactly this - as in same ledger-based, local-first/only etc. decisions - only difference really was that I never got as far as visualisations, and faffed about with PDF->ledger instead until I lost motivation for the project.
Very nice to see! I look forward to trying this out properly soon. I recommend not bothering with PDFs, OCR errors however occasional ruin it - it's worse if they're rare in a way, since you come to trust it. If you really do want it, extract the actual text layer instead. There won't always be one, tables will be a mess, and sometimes there'll be a lot of nonsense - but at least the correct content that is there will be £74.97 or whatever, not misread as £24.87.