GST + UPI Reconciliation Copilot (India)

Reconcile Indian GST invoice data with UPI transaction statements and produce audit-ready matched/unmatched reports. Use when the user asks to reconcile GST...

Perform deterministic reconciliation between GST invoice CSV data and UPI transaction CSV data. Generate four outputs: reconciled rows, GST-unmatched rows, UPI-unmatched rows, and a summary JSON.

Quick workflow

1. Confirm both input files are CSV and represent:
   • GST invoices/sales register
   • UPI collections/statement
2. Validate required intent-level fields exist (invoice id/date/total, txn date/amount/status).
3. Run:

python3 scripts/reconcile_gst_upi.py \
  --gst-csv /path/gst.csv \
  --upi-csv /path/upi.csv \
  --output-prefix /path/out/recon_2026_03 \
  --date-window-days 7

4. Read and report key metrics from *_summary.json:
   • matched rows
   • unmatched GST rows
   • unmatched UPI rows
   • reconciliation coverage %
5. Provide next actions for unmatched rows (follow-up / corrections / data cleanup).

Matching policy

• Match only UPI rows with success-like status: success, completed, captured, paid.
• Require amount match (±0.01 tolerance).
• Enforce date window (default 7 days).
• Boost confidence if invoice number or customer tokens appear in UPI note/txn_id/UTR.
• Ensure one UPI transaction maps to one invoice only.

Edge-case handling

• Ignore failed/pending/reversed UPI statuses for settlement matching.
• Preserve GST rows with empty/invalid dates as unmatched (do not force guesswork).
• Preserve UPI rows with missing amount as unmatched.
• Support flexible date formats in both files.
• Handle currency symbols and commas in amount fields.

Required outputs to share with user

Always return:

1. Reconciliation snapshot:
   • matched rows / total GST rows
   • matched amount / total GST amount
2. File paths generated:
   • *_reconciled.csv
   • *_gst_unmatched.csv
   • *_upi_unmatched.csv
   • *_summary.json
3. Priority action items:
   • high-value unmatched GST invoices
   • suspicious UPI rows (success + high amount + no invoice)

References

• Read references/csv-schemas.md for accepted columns and alias mapping.