Build an SME Trade Invoice Reconciliation and Document Correction Pipeline — implementation

## Build
Reconcile commercial invoices against Singapore GST requirements, purchase orders, and proof-of-delivery receipts. Detect price drift and missing tax registration details to generate structured correction notices.

## User
SME Treasury Accountant

## Workflow
1. Extract structured line items, seller UEN, and GST breakdown from normalized SME commercial invoices.
2. Cross-check unit quantities and extended totals against matching delivery orders and purchase orders.
3. Calculate GST computation discrepancies and flag missing or malformed Singapore tax registration numbers.
4. Generate an itemized invoice reconciliation summary and formal vendor document correction notice.

## Interface
A split-screen web reconciliation ledger displaying invoice fields, matched purchase order lines, discrepancy badges, and an editable correction draft.

## Data
Generate synthetic JSON invoice datasets with deliberate calculation errors and mock Singapore UEN numbers in a local test suite.
Starter examples below are synthetic. They are not downloaded records.

## Source
[Singapore open data API](https://guide.data.gov.sg/developer-guide/api-overview)
Utilize data.gov.sg dataset search schema specifications to format tabular financial reconciliation records.

## Tools
pydantic-ai, fastapi, pytest.
Pydantic AI guarantees type-strict schema parsing and deterministic validation for invoice arithmetic and tax IDs.
Hosted general LLM with strong mathematical and table extraction reliability.

## Deliverables
Submit runnable source, pinned dependencies, setup commands, a working interface, example outputs, and a recorded end-to-end demo. Include tests and an action log.

## Acceptance
- Identifies unit price variances exceeding pre-set SME tolerance thresholds.
- Verifies that 9% Singapore GST is applied correctly across taxable sub-totals.
- Synthesizes structured correction notices citing specific mismatched line items.
- Maintains audit logs of human approvals for corrected invoice ledgers.

## Starter Examples
```json
[
  {
    "input": {
      "invoice_id": "INV-SG-2025-01",
      "seller_uen": "201812345M",
      "subtotal_amount": 1000,
      "gst_rate": 0.09,
      "stated_gst": 90,
      "stated_total": 1090,
      "po_reference": "PO-8812",
      "po_expected_total": 1090,
      "po_items": [
        {
          "item_id": "IT-1",
          "qty": 10,
          "unit_price": 100
        }
      ]
    },
    "expected_output": {
      "status": "reconciled",
      "math_discrepancy": 0,
      "gst_discrepancy": 0,
      "line_item_matches": true,
      "action_required": "ready_for_payment_batching",
      "correction_notice": null
    }
  },
  {
    "input": {
      "invoice_id": "INV-SG-2025-02",
      "seller_uen": "INVALID_UEN",
      "subtotal_amount": 5000,
      "gst_rate": 0.09,
      "stated_gst": 400,
      "stated_total": 5400,
      "po_reference": "PO-9910",
      "po_expected_total": 5450,
      "po_items": [
        {
          "item_id": "IT-2",
          "qty": 50,
          "unit_price": 100
        }
      ]
    },
    "expected_output": {
      "status": "discrepancy_flagged",
      "math_discrepancy": 50,
      "gst_discrepancy": 50,
      "line_item_matches": false,
      "action_required": "vendor_correction_requested",
      "correction_notice": {
        "invoice_id": "INV-SG-2025-02",
        "issues": [
          "Stated GST of 400.00 does not match calculated 9% GST of 450.00",
          "Seller UEN format is malformed",
          "Stated total 5400.00 differs from expected PO total 5450.00"
        ]
      }
    }
  }
]
```

## Safety
Executes accounting reconciliation and draft generation only; does not disburse bank payments.
Use mock external writes. Require human review before any real deployment.