Veridex — Structured Data Extraction

Welcome to Veridex

Define extraction criteria, upload documents and drawings, and surface clean structured data with full gap reporting — ready to export or integrate.

0

Profiles

0

Files scanned

0

Fields extracted

0

Gaps found

Recent activity

◎

No activity yet

Upload your first file to get started

Loading projects…

◫

No profile selected

Create a new profile or select an existing one above

Kind

Drawings Data sheets 3D models

Project

Drawing intake

Survey → Extraction Criteria → Upload & Scan

⊕

Select a package to enable uploads

Supported formats appear after package selection

Location

Select a project to choose where to upload.

Project

Package

Show all jobs

⊞

Select a package to view results

Pick a project and a package above. Results appear once extraction has run.

Scope

Project

Package

⊞

Select a scope to view gaps

Pick a project and a package (or just a project) above to load the gap report.

Choose two revisions

Pick the older revision on the left, newer on the right.

From

To

Review the diff

Rev A · From

Rev B · To

Project

Select a project to view its documents.

📄

Select a document on the left to preview it here.

Appearance

Theme

Choose a light or dark appearance, or follow your device setting automatically.

Account

Change password

Choose a strong password. Minimum 8 characters. You will stay signed in on this device.

Extraction behaviour

Confidence threshold

Results below this level are flagged for manual review in the gap report

80%

How confidence scoring works

When Veridex extracts a value from a document, the AI simultaneously estimates how certain it is about that result. This is the confidence score — a number from 0 to 100 — reported alongside every extracted value.

90–100%

High confidence

Clearly printed, unambiguous value in an expected location. No review needed.

70–89%

Medium confidence

Value found but may be partially obscured, small font, or in an unusual position. Worth a spot check.

Below 70%

Low confidence

Handwritten, poor scan quality, or ambiguous. Review recommended before using.

What affects confidence scores

+ Clean digital PDF — text is machine-readable, not scanned. Consistently scores 90–99%.

+ Standard template — field is in an expected title block location matching common drawing standards.

+ Format match — extracted value satisfies a defined format pattern, confirming the right field was found.

− Scanned image — OCR introduces uncertainty. Low DPI scans score significantly lower.

− Handwritten content — significantly reduces confidence, especially for names and codes.

− Non-standard layout — field found but in an unexpected position or using an unusual label.

What the threshold setting does

Your threshold (currently 80%) is the cutoff between accepted and flagged results. Any extraction below this score appears in the Gap Report as a low confidence gap — the value is still shown in the Results table, but it is marked for human review before being trusted.

Recommended threshold by use case

Use case	Recommended threshold	Reason
High-stakes documents (contracts, certifications)	90%	Flag anything uncertain — accuracy is critical
Engineering drawings, standard templates	80%	Good balance — catches real issues without over-flagging
Mixed quality scans, legacy documents	70%	Avoids excessive false flags on old or poor quality material
Bulk processing, low-risk data	60%	Maximise throughput, review only clear failures

Note: confidence scores are self-reported by the AI model and are indicative rather than mathematically precise. Use them as a triage guide, not an absolute measure of accuracy. When in doubt, always verify against the source document.

File handling

What happens to uploaded files after extraction completes

Data management

Clear all data

Remove all profiles and results from this session

—

Backend

—

Jobs queued

—

Jobs complete

—

Tags extracted

Recent extraction jobs

Loading...

Users

Loading...

Invite a new user

User receives a magic link by email. They click the link to activate their account and set their password.

System configuration

Loading...

Enum libraries

Reusable enum value lists. Bind to fields in Extraction criteria.

Loading...

Edit library

Slug snake_case identifier used internally

Display name user-facing label

Description optional

Values press Enter to add

Used by

—

Source precedence

Decide which source wins when drawing, datasheet and IFC report different values for the same attribute.

Project

Select a project to view or edit precedence rules.

Class link source

Decide which source supplies a tag's CFIHOS class, and in what order. The first enabled source that has a class for a tag wins. New sources appear here disabled until you place them.

Project

Select a project to view or edit its class link source precedence.

Guided steps

Customise a page's guided checklist for this project. Reorder steps, show or hide them, edit the wording, and pick the completion check each step uses. Pages with no custom setup use the built-in default.

Project Page

Select a project and a page to view or edit its guided steps.

Revision scheme

How this project's drawing revisions are ordered. Used to detect a new revision of an existing drawing on upload.

Project

Select a project to view or edit its revision scheme.

Workflow

The organisation-default workflow. Individual projects can override this in their own Settings.

Schema

Explore tables, columns, and foreign keys

Schema visualiser will mount here.

Error log

Failed jobs and system errors

Loading...

Organisations

Super-admin view. Creates and manages tenant organisations.

Loading...

Visible columns