Thanks, everyone! These suggestions have been great! I am probably going to start with a 3rd party (Either Surya cloud or Microsoft) and then maybe graduate to surya or Tika if costs become too crazy.
Thanks again! (And keep the suggestions coming as this is a great list!)
adding one more option here, with obvious bias: we built DocuShell.
The gap we’re trying to solve is between raw libraries like pdfminer.six / Tabula and bigger cloud Document AI services. The libraries are great when the PDF shape is predictable, but once you need queues, retries, validation, artifact downloads, rate limits, cleanup, and consistent JSON/Markdown output, you end up building a lot around the parser.
DocuShell is meant to be that managed layer: upload a PDF through an API, get normalized JSON / Markdown / text / table-friendly artifacts back, with strict validation and short-lived/ephemeral file handling. The idea is not “we wrote the magical perfect PDF parser”, no such thing exists, but rather: use vetted extraction engines behind one API contract, with security and retention controls built in.
I’d still use:
pdfminer.six / pdf2txt.py for simple text extractionWould be very interested in feedback from folks here, especially on the Elixir integration side, since the API-first shape should make it easy to wrap with Req.
I’ve found combining Elixir with Tabula (via Java interop) pretty handy for extracting tables from PDFs. Also, the PDF spec is a fascinating read if you’re feeling adventurous!