Vision reads your documents the way a trained analyst would — pulling structured data, identifying document types, and continuously refining its understanding of your specific formats.
Core Technology
By Vision
CogniVision is Vision's purpose-built document intelligence engine. It combines deep-learning OCR, spatial layout analysis, and adaptive model training to understand documents the way humans do — recognising context, structure, and meaning, not just characters on a page.
Understands page layout, not just characters — columns, tables, headers, and the relationships between them.
Each correction refines CogniVision's understanding. Accuracy compounds across every document your team reviews.
PDFs, scans, and photos — CogniVision handles every input type without separate pipelines.
Isolated processing environments. Your documents never train a shared model or leave your configured data region.
What Vision does
CogniVision goes beyond character recognition — it understands your documents structurally and semantically, so your team receives clean, usable data.
CogniVision reads printed text and table structures across PDFs, scans, and photos — including low-quality or skewed field captures where conventional OCR breaks down.
CogniVision identifies document types before extraction begins — invoices, contracts, identity documents, reports — and routes each one to the right pipeline without manual triage.
Upload labelled examples and CogniVision learns the patterns specific to your industry and document formats. No machine learning knowledge required — just your documents and your field names.
Results come back as clean JSON or CSV, ready to push directly into your ERP, CRM, or internal workflow — no reformatting, no fragile scripting in between.
Non-technical users get a drag-and-drop web interface. Developers get a documented REST API for full programmatic access. Both surfaces expose the complete feature set.
Every document is processed in an isolated environment. Nothing crosses tenants, nothing trains a shared model, and every extraction is logged in a complete, immutable audit trail.
Process
Four steps to a production-ready extraction pipeline. No data scientists, no infrastructure work.
A workspace organises your document activity. Within it, collections define which fields to extract per document type — one collection per document category.
Upload 10–20 representative documents. Variety matters: a mix of real-world examples — different layouts, quality levels, edge cases — gives CogniVision what it needs to generalise well.
Draw bounding boxes around values and assign them to fields. Each label teaches CogniVision the relationship between your field names and where those values appear on the page.
Training completes in minutes. Test the model against unseen documents, review its confidence scores, and deploy. From that point, live documents come in — structured data comes out.
Custom model training
Generic models handle common formats well. But every organisation has its own layouts, field names, and edge cases. CogniVision custom training means the model understands your documents specifically — not documents in general.
Upload representative documents, highlight the values that matter, and CogniVision handles the learning. No regex, no rigid field coordinates to maintain as your suppliers change their templates.
Every correction your team makes during review feeds back into the model. CogniVision gets sharper as your document volume grows — without any additional configuration work.
Every training run is versioned and stored. If a new model underperforms on a specific document class, review and compare previous versions in one click without losing any data or history.
Industries
Wherever unstructured documents slow your operations, CogniVision removes the bottleneck.
Automate invoice capture, line-item extraction, and GL coding. Remove the manual steps that slow your payables cycle.
Process applications, permits, and declarations at scale with a complete, auditable record of every extraction decision.
Digitise referrals, discharge summaries, and consent forms. Output integrates directly with clinical systems — no costly middleware required.
Extract data from waybills, packing slips, and customs declarations — including photos taken in the field under poor lighting conditions.
Pricing
Pay for what you process. No per-seat fees, no hidden charges — just clean pricing that scales with your volume.
| 0 – 2,500 pages | $0.20 |
| 2,501 – 10,000 | $0.15 |
| 10,001 – 30,000 | $0.10 |
| 30,001+ | $0.05 |
Estimate your cost with Business tier volume pricing.
Get in touch and we'll set up your workspace and show you how Vision works with your own documents.