A synthetic finance-document factory for testing AI finance-operations agents: invoices, purchase orders, GRNs, bank and credit-card statements, receipts, tax docs, emails and Slack events — driven by a scenario engine with deliberate, ground-truth-labeled mutations.
GST/US invoices, credit notes, POs, GRNs, statements, 1099s, W-9s
25 benchmark scenarios with machine-readable expected outcomes
Identity, tax, matching, bank and document-quality corruptions
Realistic .eml messages and Slack events tied to each document
Scenario Registry
-> Synthetic Data Generator (seeded RNG, IN/US)
-> Mutation Library (identity / tax / matching / bank / document)
-> Validation Engine (arithmetic, tax, 3-way match)
-> Document Renderer (PDF / PNG / CSV / XLSX / JSON / EML / TXT)
-> API Layer (/generate, /generate/batch, /scenario, /document/:id)
-> Ground truth: expected_exceptions, expected_actions,
human_approval_required, dangerous_actions_that_must_not_be_automatic
| Endpoint | Purpose |
|---|---|
POST /generate |
Generate a single document in any export format |
POST /generate/batch |
Generate many documents/scenario runs at once (max 200) |
POST /scenario |
Run a named scenario end-to-end and get the full ground-truth bundle |
GET /scenario |
List all available scenarios |
GET /scenario/:run_id |
Retrieve a previously run scenario bundle |
GET /document/:test_id?format=pdf |
Retrieve/re-render a previously generated document |
GET /health |
Health check |
Note: Every generated artifact carries a visible
DUMMY / SYNTHETIC TEST DOCUMENT marker
and synthetic: true metadata. Nothing
produced here is a real financial record.