Three AI products. One intelligence layer.

AI Product 01

Data Extraction

Invoices, purchase documents, engineering drawings, bills of material, customs paperwork and 500-page scanned manuals — read, classified and converted into structured data your system can post.

Most enterprise data entry is a person reading one document and typing it into another screen. Data Extraction removes that step. The AI identifies the document type, locates the fields that matter, extracts them into a structured model, and hands them to your application for review or direct posting.

• PDF and Excel ingestion including scanned and image-based documents
• Automatic classification of document type before extraction begins
• Page-level source reference so every extracted value can be traced back to where it came from
• Validation rules against your masters — vendor codes, item codes, ports, charge heads
• Review and correction screen before anything is committed to your database
• Full audit trail of uploaded files, processing status, extracted results and final output

Typical document types

What you get back

• Structured JSON that maps directly to your transaction tables
• Standard Excel outputs — material data sheets, deviation sheets, drawing data sheets
• Direct posting into the ERP screen, pre-filled and ready for approval

AI Product 03

Ask AI

One assistant button that can be dropped into any transaction form. It reads the record on screen, explains what it is, who touched it, why it is stuck and what to do next — in plain English.

Ask AI is configuration-driven. Onboarding a new screen means adding rows to a configuration table that define which queries apply to that form — no code change to the service, and no new deployment. Key values are always passed as parameters, and the AI credentials never leave the server.

• Drop-in button on any form — three lines of markup per screen
• Config-table onboarding — add a new form by adding rows, not by changing code
• Multiple query sets per form — header, lines, audit trail, error log
• Parameterised queries only, so the key value can never be injected
• Server-side credentials — the AI API key is never exposed to the browser
• Read-only by design — the feature only ever reads data, never writes it

How it behaves

1. Reads the recordPicks up the key value from the form — a text field or a selected grid row.
2. Runs the configured queriesThe shared service looks up which SQL applies to that form and executes it with the key as a parameter.
3. Interprets the resultThe combined data is sent to the selected AI model along with the form context.
4. Explains it in the popupWhat the record is, who created or changed it, the likely root cause and the suggested next step.

Questions users actually ask

How it works

From document to decision

The same pipeline underpins all three products — only the inputs and outputs change.

1. IngestFiles arrive through a secure web form, a watched folder or an API call.
2. Read & classifyThe AI identifies the document type and the requirement data inside it.
3. Retrieve referencesRelevant catalogue pages, standards and historical records are located using a retrieval index with page-level references.
4. Compare & validateRequirements are compared against the reference information and any mismatch or deviation is identified.
5. GenerateStructured JSON is produced first, then the final output — a posted record, a standard Excel sheet, a report or a plain-English answer.
6. Review & exportA human reviews and corrects where required, then the final export and audit record are written.