Introduction

Inferio is an OCR pipeline powered by large language models. You upload a document; we return structured JSON with per-field confidence scores. What you build on top of that is your call — accounting integration, AP automation, KYC onboarding, contract intake.

What Inferio handles

• Layout-aware extraction — Claude Vision reasons about document structure, not just pixel templates. Works on rotated phone shots, handwritten fields, noisy scans
• JP compliance — every T-number is validated live against the National Tax Agency API; archives are timestamped to meet the Electronic Bookkeeping Law
• Validation — domain-specific rules per document type (IBAN/BIC for bank statements, MRZ for KYC, date + amount cross-checks)
• Audit trail — every extraction, validation, and correction is logged with timestamps, exportable per request
• Confidence scoring — every field gets a 0.0–1.0 score; anything below 0.75 surfaces in the correction UI before sync

What you handle

• Auth — generate API tokens in the dashboard; ship them server-side, scoped per environment (test/live)
• Document choice — pick from seven built-in document types or define a custom JSON schema in the dashboard
• Webhook receiver — implement an HTTP endpoint that consumes ocr.document.completed and writes to your downstream system
• Operator workflow — decide when to require human review (the default 0.75 confidence threshold is configurable per merchant)

Architecture overview

   ┌──────────┐
   │  Your    │
   │ system   │
   └────┬─────┘
        │ POST /api/v1/ocr/{...}
        ▼
   ┌────────────────────────────────────┐
   │           Inferio                   │
   │  ┌─────────┐  ┌──────────────────┐ │
   │  │Upload   │→ │Pre-process       │ │
   │  │session  │  │(deskew, denoise) │ │
   │  └─────────┘  └────────┬─────────┘ │
   │                        ▼           │
   │             ┌─────────────────────┐│
   │             │AI extraction        ││
   │             │(Claude Vision)      ││
   │             └────────┬────────────┘│
   │                      ▼             │
   │             ┌─────────────────────┐│
   │             │Validation           ││
   │             │(T-number, IBAN, …)  ││
   │             └────────┬────────────┘│
   └──────────────────────┼─────────────┘
                          │ signed webhook
                          ▼
                  ┌───────────────┐
                  │ Your webhook  │
                  │ receiver      │
                  └───────────────┘

The three integration surfaces

Next