Mahal Systems CORD
Fits receipt and expense capture agents. Borrow its two-level label scheme (five superclasses, 42 subclasses) and its line-group annotations as a template for structured extraction targets.
What it asks the agent to do
- extract menu items, prices, subtotal and total from a receipt photo
- group extracted fields into line items
- parse a receipt after OCR into a structured record
What to borrow from it
case templatebaseline
Caveats
From 2019 and small (1,000 Indonesian receipts, 100 in test); the repo ships no scorer, so field-level scores across papers are not directly comparable; essentially solved by current document models and useful mostly as a smoke test.
Details
| Workloads | documents, retail |
|---|---|
| Interaction | single turn |
| Environment | static |
| End state checked | No |
| Tools | none |
| Grading | programmatic check |
| Size | 1,000 |
| Data | real public |
| Personal data | possible |
| License | CC-BY-4.0 |
| Harness | none |
| Status | frozen |