Introducing Contexia™ and DeepSight™

Overview

Metis is adding two solutions to its suite for regulated enterprises: Contexia™ and DeepSight™. Contexia helps teams understand requirements, tables and business flowcharts across complex document sets, with every answer linked back to its source. DeepSight turns scanned and unstructured documents into checked, source-grounded data that teams can review before use.

Both solutions run on the Metis ARX private core, so sensitive content stays within the customer's own environment.

Why these solutions matter

Regulated organisations run on documents: business requirement documents (BRDs), functional specifications, application forms, invoices, circulars and tender packs. Much of this still gets read, keyed and cross-checked by hand. Generic AI tools read plain text and miss the structure of a table or a flow diagram. They also cannot be used with confidential documents if the content has to leave the organisation.

Contexia and DeepSight address both problems. They read the documents as they are laid out, they show where each answer came from, and they run where the data already lives.

Contexia™: Visual RAG

Standard retrieval-augmented generation (RAG) works on extracted text. Contexia reads structure, relationships, business intent, and layout and design. It parses headers, tables and flow diagrams in BRDs, functional specs and business flowcharts, and builds a semantic model of every requirement. Each result links to its source.

What teams can do with it

Ask questions across a large document set and get answers tied to the exact source passage.

Understand how requirements relate to each other, including those held in tables and flow diagrams.

Generate test designs for end-to-end, negative and edge-case paths, with every test traced to its source requirement for audit readiness.

Produce compliance and bug reports in the company's existing document standards, with layout and format preserved.

Contexia also powers Metis UAT Automation. Business testers configure scenarios in a no-code interface, tests run unattended and in parallel, screenshots and video are captured as evidence, and critical sign-offs stay human-controlled.

DeepSight™: source-grounded OCR

DeepSight uses a vision-language engine to read scanned forms, stamps, handwriting and unstructured documents in English, Traditional Chinese and Simplified Chinese. It turns them into structured data that staff review before it is used.

Typical performance (configuration-dependent)

Over 98% extraction accuracy on printed text and over 94% on handwritten text.

About one page per second.

400 scanned applications processed in under 24 hours.

How it keeps the output reliable

A rules engine checks the extracted fields against the organisation's own criteria.

Missing items are flagged, and low-confidence fields (below 90%) are highlighted for human review.

Supervisors are notified automatically on exceptions or eligibility mismatches.

Answers and drafts are generated only from retrieved sources and carry page and paragraph citations in English and Chinese. If no source exists, the system says so and does not guess.

DeepSight accepts PDF, image, Excel, JSON and CSV inputs, and outputs text, CSV, JSON, Markdown or Excel.

Built on the Metis ARX private core

Both solutions inherit the controls of Metis ARX:

Air-gapped deployment on on-premises enterprise hardware, with no external calls during inference.

Document vectors and model parameters stored locally.

Role-based access control, so people see only what they are authorised to see.

An immutable audit trail covering queries, outputs and document access.

Certification to ISO/IEC 27001 and ISO/IEC 42001.

Where an organisation also uses Metis AERIX, policy routing keeps sensitive content on ARX and sends only public-safe work to approved cloud models.

A working example

A bank receives hundreds of scanned account applications. DeepSight extracts the fields, including handwritten entries, and the rules engine checks them against eligibility criteria. Applications that pass are marked for approval, incomplete ones are flagged with the missing item, and uncertain fields go to a reviewer. Meanwhile, the bank's IT team gives Contexia the BRD for a system change. Contexia produces traceable test cases and a draft test report, which testers review and sign off.

Availability

Contexia™ and DeepSight™ are available as part of the Metis solution suite. They run on the Metis ARX private core, and a typical implementation takes about 16 weeks from hardware setup to go-live. Organisations can begin with a pilot and add modules as they scale.

To learn more or arrange a demo, contact enquiry@metis.info.

About Metis

Metis is a private and hybrid AI platform for banks and other regulated enterprises.