Repair the file first
Broken flate streams and corrupted compression are re-encoded with pdf-lib before parsing begins. A readable file in, every time.
flate stream broken → re-encoded · opened clean
Numbers you can hand
to the revenue service.
The financial engine reads bank and portfolio statements, rebuilds every transaction into one canonical schema, and refuses to export a ledger that doesn’t reconcile. Filing-grade, not best-effort.
Same columns, whatever the bank.
A Capitee statement and an Investec one look nothing alike going in. What comes out is the same six columns every time: date, description, category, debit, credit, running balance.
| Date | Description | Category | Debit | Credit | Balance |
|---|---|---|---|---|---|
| 01 Sep | Salary · client retainer | Income | - | 48 200,00 | 112 480,14 |
| 03 Sep | Card · Shoprite | Debits & Expenses | 1 240,50 | - | 111 239,64 |
| 05 Sep | Debit order · rent | Debits & Expenses | 14 500,00 | - | 96 739,64 |
| 07 Sep | Bank charges · VAT 15% | Bank Charges & Taxes | 46,00 | - | 96 693,64 |
How a statement becomes proof.
From a skewed photo of a statement to a reconciled workbook, every file runs the same disciplined route. Nothing is guessed twice, and every balance is checked on the way out.
Name the bank, fall through six strategies
Fourteen signatures are matched against headers and footers and scored. Then the file drops through a self-healing chain until it reconciles.
Text layer → OCR → Normalized AI → Vision AI → Hybrid → Spatial rescue
Pull every line
Dedicated extractors for the major banks; spatial-alignment rescue rebuilds tables from coordinates when the text layer lies.
x 112.4 · y 480.9 → row 41 of 87 rebuilt from coordinates
Prove the numbers
Rows become canonical transactions, and the running balance is re-derived line by line. If it does not reconcile, it is flagged, never silently shipped.
Σ running balance = closing balance verified
Five deterministic layers between your numbers and a wrong total.
Reconciliation here isn’t a model guessing. Each layer is plain arithmetic that re-adds the ledger and blocks the export when a single check fails.
Opening to closing, re-added
Each line’s balance is recomputed from the last. If it doesn’t match the bank, the row is flagged before it reaches the sheet.
Never both, never neither
A transaction gets exactly one side, cross-checked against the balance movement.
The R10k sanity net
An amount far outside a row’s context is surfaced for review instead of silently categorised.
Same line, counted once
Re-imports and overlapping pages are fingerprinted so a single debit never doubles the expense.
It will not ship a number it can’t re-derive
The whole ledger must balance from opening to closing before export, the check that makes output filing-grade.
Built around SARS VAT 501.
Over a hundred and twenty categorisation rules run in English and Afrikaans. They match how a line really reads on a local statement, BANKKOSTE and all, and land it in the right box.
Anything it can’t match goes to a human to review, not a guess. And each client’s rules stay separate, so one book can’t quietly change another.
Nine-column VAT schema
Standard suppliesZero-ratedExemptNon-purchasesImports VATAdjustments
15%
Output & input VAT, with the 115 registration gate handled.
R7 000
De minimis asset rule; below it, immediate write-off logic.
PAYE · UIF
Payroll lines recognised alongside salary templates.
Eskom → Santam
Local merchants, vendors and debit-order patterns pre-known.
It already knows fourteen banks.
A phone photo of a statement reads as cleanly as a pristine PDF. These are the banks and fund managers the engine knows by signature.
FN
FNB
Transactional & business
CA
Capitec
Transactional
NE
Nedbank
Transactional & business
AB
ABSA
Transactional & business
SB
Standard Bank
Transactional & business
TY
TymeBank
Transactional
DI
Discovery Bank
Transactional
IN
Investec
Private & wealth
BV
Bidvest
Corporate card
AF
African Bank
Transactional
IC
ICICI
Transactional (India)
OM
Old Mutual
Portfolio & unit trusts
SA
Sanlam
Portfolio & unit trusts
AG
Allan Gray
Portfolio & unit trusts