# Iris: Visual data extraction

Understand line / Stop 01. Turns chaotic document streams into reliable, auditable data.

Iris turns PDFs, scans, images and Word documents into validated, structured JSON, with a confidence score and source excerpt on every value.

## What It Does

- **Multi-Format Extractors**: PDFs, Word, images, spreadsheets and handwritten forms, preserved with layout and table fidelity.
- **Dynamic Schema Prompts**: Define the fields you need and Iris reasons to find them across variable layouts, with no retraining for every template.
- **Confidence and Provenance**: Each extracted value shows a confidence score and source excerpt for quick QA and audit.
- **Structured JSON Output**: Clean, schema-aligned payloads ready for databases, analytics or compliance pipelines.

## Control and Audit

- **Guardrails**: Rule validation enforces thresholds, such as safety limits, on every extracted value.
- **Where a person decides**: Exceptions are surfaced to human review, so staff stop copy-pasting and focus on the cases that need them.
- **What gets logged**: Provenance links every value back to its source extract, so data stays auditable for regulatory workflows.

Included on every line: Tenant isolation, Full audit trail, POPIA and GDPR, Model choice. Praxis is the agent framework in VectorMind. Agents plan, call tools and validate their results, and every step is logged with its tool inputs and outputs.

## Where It Runs

- **Deployment**: Managed, VNet or air-gapped installs, tenant-aware, to meet data sovereignty needs.
- **Platform**: Cloud, private cloud or sovereign on-premise environments, with tenant isolation.
- **POPIA**: CohesionX platforms never train on your sensitive information, and follow strict access controls, encryption and compliance protocols, including POPIA and GDPR.

## Specs

- **Inputs**: PDF, Word, Images, Spreadsheets, Scanned forms, Handwritten forms
- **Outputs**: Schema-aligned JSON, Confidence scores, Source excerpts
- **Integrations**: Databases, Dashboards and analytics, Compliance pipelines
- **Deployment**: Managed, VNet, Air-gapped

## On the Line

Iris is on the Understand line with [ConversationIQ](/products/conversationiq.md), [Runestone](/products/runestone.md).

Book a walkthrough: info@cohesionx.co.za