Skip to content
Technology

Built on DexAI's own AI technology.

DexAI is an AI-powered financial document and automation platform, not an OCR tool with an accounting skin. Here is what runs underneath.

  • 01

    Proprietary OCR

    Text extraction built for financial paperwork rather than general documents.

    • Advanced document text extraction
    • Structured data extraction
    • Financial document understanding
  • 02

    Proprietary VLM

    A vision-language model that reads the page the way a person does.

    • Understands document layout
    • Understands visual context
    • Handles complex financial documents
  • 03

    Document Intelligence

    The layer that turns reading into finished work.

    • Classification
    • Extraction
    • Validation
    • Automation
Agentic AI

Not one model. A team of specialist agents.

DexAI is not simply OCR or an accounting add-on. Each document is handled by a chain of focused AI agents, from classification to reconciliation, with a person brought in only where confidence is low.

Stage 01 of 07

Document

A document arrives by camera, upload or email forwarding and enters the pipeline with its original file attached.

The agents

Eight agents, one job each.

  • 01

    Invoice Agent

    Reads supplier invoices, capturing header fields, line items, VAT and totals.

  • 02

    Receipt Agent

    Turns receipt photos, scans and emailed confirmations into structured expense records.

  • 03

    Classification Agent

    Identifies the document type on arrival and routes it to the right specialist.

  • 04

    Extraction Agent

    Pulls supplier, date, reference, VAT and totals from the document as separate fields.

  • 05

    Validation Agent

    Checks the arithmetic and formats, flagging low-confidence fields for review.

  • 06

    Reconciliation Agent

    Pairs records with bank transactions on amount, supplier and date.

  • 07

    Reporting Agent

    Keeps expense and VAT summaries current as records are added or corrected.

  • 08

    Data Migration Agent

    Maps and moves historical records into DexAI with duplicate checks along the way.

AI training

Built for large-scale document intelligence.

DexAI's models are trained on financial documents specifically, evaluated before every release and improved continuously from review corrections.

  1. 01

    Large-scale training

    Models are trained on large volumes of financial documents rather than tuned on a handful of templates.

  2. 02

    Diverse document datasets

    Receipts, invoices, statements and claims in many layouts, languages of numbers and print qualities.

  3. 03

    Financial-document specialisation

    Training is focused on the fields that matter in bookkeeping: suppliers, dates, VAT, line items and totals.

  4. 04

    Continuous model improvement

    Corrections made during review feed back into evaluation and future training rounds.

  5. 05

    Model evaluation

    Releases are measured against held-out document sets before they reach production.

Training figures are published only once verified.

Infrastructure

Flexible AI infrastructure for real-world deployment.

DexAI's models are built to run where you need them, on CPU-only hardware or GPU-accelerated systems, without changing how the product behaves.

Inference target
Documents
Receipts, invoices
DexAI models
CPU-only
Records
Fields + confidence
Model build
Same models, optimised for either path
Scaling
Add capacity as document volume grows
Latency
Designed for low per-document latency
Portability
Deployment driven by your requirements
  • GPU flexibility

    01

    Inference is not tied to a single accelerator family.

  • CPU / GPU optimisation

    02

    Models are optimised to run on CPU-only hardware as well as GPU-accelerated systems.

  • Efficient inference

    03

    Document processing is designed to keep per-document cost and latency low.

  • Scalable deployment

    04

    Capacity grows with document volume without changing how the product works.

  • Hardware flexibility

    05

    Deployment choices are driven by your requirements, not by the model's hardware demands.

Architecture

From document to business system, stage by stage.

Every document follows the same path. Select a stage to see what happens there.

Stage 01 of 08

Documents

Receipts, invoices, statements and claims arrive as photos, PDFs and emails.

Get started

Ready to automate your finances?

Spend less time processing documents and more time running your business.