Extraction Models
How Fortifiers uses AI to extract data from documents.
Hybrid Extraction Engine
Fortifiers uses configured hosted extraction and deterministic validation for supported field formats. Results can be incomplete or incorrect and must be checked against the original document.
1. OCR
Optical Character Recognition
Converts images and scanned documents into machine-readable text.
- Configured OCR services may vary by environment
- Review extracted text against the original scan; no general accuracy percentage is promised
2. LLM Extraction
Large Language Models
Suggests candidate fields from document text and configured extraction guidance.
- Configured hosted AI extraction
- Combines document context with configured field guidance
3. Pattern Matching
Regex & Rules
Deterministic extraction for structured fields like dates, amounts, and IDs.
- Deterministic validation for supported field formats
- Validates LLM outputs
Extraction Pipeline
- 1
Document Upload
File is received and classified by type.
- 2
OCR Processing
Text is extracted from images/PDFs.
- 3
Layout Analysis
Document structure is identified (tables, headers).
- 4
LLM Extraction
AI model extracts key fields based on document type.
- 5
Validation & Review
Data is validated against rules; low confidence flagged.
Review requirements
Model confidence is a review aid
A score is not a calibrated accuracy guarantee, and no score automatically approves extracted data.
Check every field against the source
OCR and extraction can omit, misread, or infer details. A person should verify values before using them in a quote or business decision.
Extraction Configuration
Admins can configure extraction guidance for supported document types and review model output before using it.
- Define expected fields and document guidance
- Compare extracted fields with the source document before using the output