Three parsers turn one page into three different indexes.
Page 7 of a two-column product handbook goes through three LangChain PDF loaders, as in lesson 11.4. Pick a loader, then hover or focus an output Document to see which part of the page it came from.
handbook.pdf, page 7
4. Plans and pricing
Every plan includes DDoS protection, daily snapshots and a dedicated IPv4 address. Outbound traffic above the included amount is billed at $0.01 per GB.
| Plan | vCPU | Memory | Traffic | Price / month |
|---|---|---|---|---|
| Starter | 1 | 2 GB | 1 TB | $6 |
| Business | 4 | 16 GB | 5 TB | $48 |
| Scale | 16 | 64 GB | 20 TB | $190 |
5. Refunds
Annual plans can be refunded; monthly plans cannot. Refunds go back to the original payment method within 5 to 10 business days.
- Within 30 days: full refund if outbound traffic is under 100 GB.
- After day 30: unused whole months, minus a 15% early-termination fee.
- SLA and promotional credits are never paid out as cash.
Output of
struck: header or footer text inside the contentmarked: a line-break hyphen left in a word
- Documents from this page
- Header and footer
- Table rows
- Section known
The lab’s metadata schema, filled by this loader
| Field | Used later for | Status | How |
|---|
Outputs are written to show each loader’s default behaviour on a two-column page with a table: one Document per page for PyPDF and PyMuPDF, one per element for Unstructured in mode="elements" with the hi_res strategy (the fast strategy finds no tables and fewer headers). Element ids are shortened to e4 and e5; real ones are hashes. Exact text varies by PDF producer, so run lesson 11.4’s notebook on your own files before you pick a parser.