The deterministic reliability layer for finance data
Messy bank exports and invoices in, validated structured data out — produced by rule-based parsers, never a language model. The same file always yields the same result, and every value comes with a reason.
What APODEXA does
It reads the finance files you already have and returns a clean, validated result you can put in front of an auditor — with the checks it ran and the run's provenance (source file, parser, and transformations) attached.
Five deterministic tools cover the common jobs:
- Detect file type — identify what a file actually is before parsing it.
- Repair CSV — fix broken delimiters, encodings and headers without dropping rows.
- Normalize bank CSV — turn a bank/statement export into a clean canonical ledger.
- Extract invoice — pull structured fields out of a PDF or text invoice.
- Validate invoice — check an invoice's totals and structure add up.
On top of the tools, running-balance reconciliation checks that each balance follows from the row before it — so a missing, altered or duplicated transaction is caught rather than passed. Validated ledgers export to a Xero-compatible CSV for your accounting workflow.
Every response is an envelope you can trust
Each tool returns the same structured envelope, so results are machine-checkable and reproducible:
decision
An explicit accept, review or reject — ambiguous input is rejected or routed to review, never guessed.
output_hash
A hash of the canonical output. Re-run the same file and you get the same hash, so any change is visible in a diff.
provenance
Every run records where its data came from: source file, parser, mapping and each transformation applied — with a source-row reference for every normalized bank row.
confidence
A heuristic score per field, surfaced honestly — low-confidence fields go to the review queue instead of being silently filled.
See it on a real file
Run a sample bank export through the no-signup demo, or read the API reference.