Today I scanned the service log of my El Camino going back to 1998 (I bought the car in 1996, dunno why it took me so long to start the service log). It was handwritten, of course, as I serviced and fueled the car. Then I asked Google Docs to OCR (that’s a verb now) the scan into text. It did a shockingly good job, considering it was handwritten.
Still, the OCR’d text was a mess. I next asked Claude Sonnet to look at the text and make some sense of it, exporting a tab-delimited table that could put into a spreadsheet. That worked pretty well, too, especially as Claude was clever enough that he would interpolate obscure odometer readings to organize everything properly (odometer readings only go up).
Finally I proofread the whole thing and printed it into a new service log as the original was getting kind of ratty.
Workflow:
1. Copy pages from the log one by one (enlarge while copying) (copies can be seen above);
2. Scan copied pages into a multi-page PDF;
3. Upload to Google Drive;
4. In Google Drive, open with Google Docs for automatic OCR;
5. Copy and paste Google Doc into text file; looks like garbage;
6. Ask Claude to fix it up into a proper spreadsheet;
7. Proofread spreadsheet;
8. Print results.
Google OCR is insanely good, and this would have been an extremely tedious project without Claude Sonnet.
Probably within six to twelve months the workflow above will seem kind of quaint, as by then Claude Sonnet will be able to do the entire thing for me once I copy the pages.
