Skip to Content
Welcome

Welcome to Inferio

Inferio is an AI-powered OCR platform that turns invoices, receipts, contracts, bank statements, and KYC documents into validated, structured JSON. Japan-compliant out of the box — every T-number is validated against the National Tax Agency public API, every archive is timestamped to satisfy the Electronic Bookkeeping Law.

You upload a PDF, image, or scan; we return structured fields with per-field confidence scores. Webhook fires the moment extraction lands. No prompt engineering, no per-template tuning.

Start here

Process your first invoice in ~5 minutes — create an upload session, POST a PDF, watch the webhook fire with extracted JSON.

Quickstart

The mental model behind requests, upload sessions, documents, and the confidence-driven correction loop — read this once and the rest clicks into place.

Concepts

Endpoint-by-endpoint specs, request/response schemas, and error codes for the /api/v1/ocr/* surface.

API reference

Wire ocr.document.completed into your downstream pipeline. HMAC signed, idempotent, dead-letter replayable.

Webhooks

What you can extract

Inferio ships with pre-tuned JSON schemas for seven document types:

  • Trade invoices (JP) — vendor, T-number, line items, 8% / 10% tax breakdown, totals, due date
  • Letters of credit — applicant, beneficiary, amount, expiry, Incoterms, 40+ fields
  • Bank statements — IBAN/BIC, transaction date, debit/credit, running balance
  • Payroll records — employee ID, base salary, deductions, net pay
  • KYC / identity documents — MRZ, full name, DOB, expiry, issuing authority
  • Receipts — merchant, items, total, tax rate, payment method
  • Custom schemas — define your own JSON shape; the same extractor adapts without retraining

How to integrate

Two integration paths share the same core engine:

  1. REST API — plain HTTP + JSON. Idempotent, retry-safe, batch-friendly. Bearer-token auth + X-Merchant-ID scope.
  2. Web dashboard — drag-drop documents, approve exceptions, audit trail. No code required.

Both paths emit the same three lifecycle events: ocr.request.created, ocr.document.processing, ocr.document.completed. Wire your downstream pipeline to the completed event and stop polling.

Trust the pipeline

  • 99%+ field-level accuracy on Japanese mixed-script invoices (Claude Vision, layout-aware)
  • <5s single-page extraction · 15–25s for a 10-page multi-page PDF
  • Per-field confidence scores — anything below 0.75 surfaces in the correction UI before sync, never silently
  • No silent retention — Inferio processes documents in-memory only; the AI provider does not retain or train on your data
  • Electronic-Bookkeeping-Law-grade audit trail for every extraction, validation, and correction
Introduction