Release Notes

• Improved webhook service performance to shorten response times

• Fixed a bug in the document representation setting that let to expired jobs

• Allow retrieval of app measurements via API.

• The order of the CSV export headers can be ordered based on app settings field order.

• Fixed a bug that caused classification jobs to go to expired

• Enhanced accuracy for Document Classification

• Fixed a bug to fail job uploads when document cannot be accessed.
• Fixed a bug that prevented job submission in case of failing job batch structure validation.
• Fixed a bug in the Classifier that led to errors when processing documents with a lot of text.

• Added new sample methods for app measurements.

• Update Panel on the statistics page displaying the performance over time grouped by field

• Fixed a bug that led to field calculations being off

• Updated the Measurer configuration page

• Fixed a bug that led to wrongly displayed worker ids for AI workers

• Added API to configure application measurements

• Fixed a bug that led to wrongly selected option
• Fixed a bug that led to a job not being created

• Fixed a bug that led to not include advices and instructions into prompt

• Fixes a bug a which lead to issues with filtering and removing tags from existing jobs

• Fixed a bug that led to extended processing time

• Added support for processing images larger than 500 KB via emails

• Fixed a bug that led to incorrect dimension of tables while editing

• Fixed a bug regarding prompt generation for older applications

• Added option to use regex validation rules when extraction via other workers fails (e.g. LLM)

• Added ability to configure the frequency and other parameters of measurements per application

• Enhance job reliability

• New pod to manage failed data uploads

• Reduced onboarding time by streamlining required steps

• Reduced project creation time

• API: Added endpoint for downloading field datasets as CSV

• Fixed a bug that led to inconsistencies when deleting application tags via API

• Fixed a bug in Database Matching that led to resulting in error for some jobs

• Introduced confidence scores for latest LLM model versions

• Performance improvements for the "Levenshtein" bounding box matching algorithm

• Document Classifier: Now propagates tags and metadata to substream projects

• Introduced on option to just classify document or page instances

• Fixed a bug in the Database lookup feature that prevents updates to the dataset that could break subsequent jobs

• Broadened support for actions shortcuts
• Enhanced date filters for statistics
• Individual application names are presented in tab title now

• Fixed a bug that led to incorrectly treating .jpg file extensions

• Fixed a bug that led to an error in rendering job output
• FIxed a bug that led rendering issue in table cells

• Fixed a bug that led to extracted values were not presented in the HITL task

• Updated behavior in HITL tasks regarding confidence scores

• Fixed a bug to allow users to edit job output If there's no possible database match

• Added field annotation count per field class

• Improved space detection for extracted numbers

• Fixed a bug that led to Job Edit View not responding

• Fixed a bug that caused empty extractions in GewA1 document type
• Fixed a bug that let to not handling .jpg extensions when converting image contents

• Various enhancements in the Database Matching feature

• Fixed a bug that let to an inconsistency in user feedback

• Fixed a bug that occasionally led to users not able to download the job JSON / CSV.

• Updated 10-K document type

• Fixed a bug that led to not showing all "To Be Labeled" KVPs and Tables in HITL task.
• Updated default KVP and OCR engine

• Added a new option for matching methods

• Performance improvements for bounding box matching task

• Fixed a bug in the Tables Headers matching algorithm
• Fixed a bug in the "Remove Annotation" action

• Document Classification: Ability to route Jobs according to app-ids. Document pages will be routed as new jobs to user defined apps.

• Enhanced table support for longer tables

• Database Lookup: Allows adding Dataset field entries within `Edit Output`. When adding annotations to "Calculated Fields", new entries will be added to the project's dataset.

• Email support: Added a setting for allowed list of email addresses for a project

• Fixed an issue where the job could get stuck
• Fixed a bug that presented the submit job button active for annotator role

• Enhanced batch upload error notification
• Usability improvement: disabled moving beyond content in canvas for high zoom values and enabled centring content on zoom in/out

• Fixed a bug that led in case of a server not reachable to AI tasks failing and the job expiring

• Introduced SuperTask feature for Calculated dataset fields. Enables manual review of candidates not meeting the initial automatic matching criteria. Provides functionality to add entirely new entries to the new super.AI Dataset.
• Added OCR combiner and KVP matching combiner. Changed default SuperTasks settings: OCR, KVP extraction and KVP matching have both the workers correctly setup by default

• Dataset matching feature can accommodate additional columns in the datasets, beyond those utilized in the matching algorithm.

• Fixed a bug in the Invoice document type's default settings.

• Table Editing enhancement: Implemented ability to drag and update table cells

• Table view is now accessible directly by double-clicking

• Changed model feedback modal to be off b default

• Fixed a bug in the LLM Table Extraction process where the model occasionally produced extraneous text. With this update, such text will now be disregarded, enhancing the reliability of table extractions based on LLM.

• Fixed a bug that led to "Multipage Tables" weren't detected

• Support for multiple databases to be taken into account in Database Lookup feature

• Various accessibility improvements
• Document Classifier: detect empty pages in a document while classifying. Can be completely empty, or pages with less than 10 words.

• Improved ux for cases no bounding box present for a value

• Fixed a bug that causes error in LLM based table extraction in case of empty documents

• Database Lookup: Allows multi-value database matching per field.
• New bounding box matching algorithm that increases out-of-the-box accuracy across all document types

‍

• New Email Attachment Scanning UI

• Fixed a bug that led to not handling failed AI predictions asexpected

• Added KVP and Table matching SuperTasks. Removed document type parametrisation, now the parameters and model selection are dependant only on app params and SuperTasks settings.

• Fixed a bug for cases where annotation bounding boxes were identical and both of them would be removed.
• Fixed a bug in GewA1 document type that causes no extractions to be shown if one field was set to inactive.
• Fixed a bug that took with respect to unknown properties from ground truth for GewA1 documents

• Improves line detection heuristic in whitespace representation

• Fixed a bug in GewA1 document type that let to filtering of redundant addresses

• Fixed a bug that led to refreshing the document page on preview
• Fixed a bug that led to the "Process myself" button not available for apps with collaborator workers

• Enhanced support for public URLs
• New Data Storage URL format

• Document Classifier: Enhanced instance classification

• New combiner method for key value pair SuperTasks; allowing for multiple models to be called in an aggregate manner

• Enhanced instance classification in multi-page documents

• Changed default value of 'Required Annotation' to 'disabled' for all fields

• Add LLM Settings in "LLM Settings" view (previously "Prompt Settings")
• Enhanced LLM table extraction for greater user control. Amongst other things, fields for description, instructions, advice, and the description and alias for custom headers have been seamlessly integrated into the prompt.
• Extended support for multiple OpenAI models as a TaskWorker
• Added 'Max Fields' as a parameter for LLM prompts

• Usability improvement: removed the need to click twice to open annotations actions popup

• Updated document types: Certificates, crash report, financial statement, utility bills

• Updated document types: Invoices, contracts

• Fixed a bug that led to not saving SuperTask parameters without HITL worker

• Fixed an issue where the correct page number of a detected stamp in multi-page documents was not being displayed. This issue only occurred when the main stamp was not on page 1.

• Enhanced options for field type "Signature"

• The HITL Settings section will now be hidden when there is no active collaborator worker in 'verify results'.

• Resolved an issue with changing of rows and columns for tables in the "Advanced Custom Preview" document type, ensuring seamless modifications.

• Document Classification: Ability to classify documents via natural language instructions (LLM based)

• Enhanced stamp detection

• Several minor usability improvements

• Fixed a bug that didn't allow to select annotation in review mode
• Fixed a bug that led to not persisting selected value in the table edit modal

• Document Classifier: LLM support
• Added context menu in table popup
• UX improvements: implemented new widget for classifier params

• Accessibility, aria labels for next, prev buttons, and annotation review actions
• Document.Extract widget works faster during panning / zooming documents
• UX improvement: Select widget search improvements

• Meta AI: added support to share models across projects within an organisation

• General UX enhancements

• Fixed processing of filenames with certain special characters

• Fixed a bug that in the API that could led to service degradation
• Fixed a bug in the Fuzzy Table Header matching

• General UX improvements in the table widget
• Added 'multi delete', and 'delete all' as well as keyboard shortcuts for value instances

• Bug fix regarding token usage estimation

• Introduced new document type: crash report

• Fixed a bug regarding email signup
• Fixed a bug that was impeding extraction when stamp extraction was disabled for form document types

• Fixed 'Invalid Data Review‘ error message when using tables

• SuperTask support for rotation task
• API: ability to exclude annotations and flatten tables in CSV files

• General UX improvements
• Enhanced error / info reporting

• Enhanced hints for prompt input

• Various fixes that lead to better extraction performance for form document type

• Fixed a bug regarding ground truth examples in form documents

• Fixed a bug regarding multi-chunk extractions

• The fuzzy matcher tasks will not be created if no annotations are present from previous steps

• Fixed a bug that caused missing extractions in long documents

• Enhanced metric support for fields without bounding boxes

• Enhanced multi-line extraction

• Added hints regarding token count for prompts

• Enhanced bounding box support for redaction

• General performance improvement for AI predictions

• Fixed a bug regarding optional annotations

• Improved bounding box support for LLM extraction

• General UX enhancements
• Option to apply redaction without taking case-sensitivity into account

• Reduced the number of partial redactions where complete redactions are desirable

• In cases where annotation bounding boxes were identical we would remove both of them. This resulted that no redaction bounding box was created in this case.

• Fixed a bug regarding app creation and SSO logins

• Fixed a bug that led to not processing jobs with inactive fields

• Enabled linking a dataset to a custom field for matching
• Added field aliases to prompt
• New version of the LLM based table extractor - supporting highly complex tables

• Performance improvement for displaying jobs with big payloads

• Fixed orientation issue for document rotation task in the rare case where an image has detected rotation and EXIF metadata
• Fixed a bug when modifying settings in an LLM project

• Updated custom fields for invoices, bank statements, license, contracts, payslips

• Fixed LLM token calculation

• New LLM options for fields

• Fixed a bug that led to datapoints getting stuck in queue upon reprocessing after an application update.

• API: Add option to get any job's history not just the latest version

• Fixed a bug to always to display high/medium/low confidence icon next to value instances
• Fixed a CSV output bug

• Fixed a bug that led in some case the fuzzy matcher not returning the best match, but the last found

• Fixed a bug with SSO login

• Work Queue: Add number of pages column and sort option
• Ability to resize table panels
• Added a guard that disregards the processing of documents with number of pages or size beyond a specified threshold

• Redaction now relies on new AI framework

• Fixed a bug in LLM prompt examples

• Ability for Admin users to edit LLM prompt components on project level

• Updated Table Identification algorithm: enhance underlying LLM prompt
• Ability to use Ground Truth samples as part of the prompt

• Table Editing: Support for vertical headers

• Table Editing: Multi-page tables highlighting support

• Fixed a bug in custom fields to correctly calculate payment terms for invoice document type
• Fixed template output rendering for custom fields

• Fixed a bug that led to an issue with detecting signatures

• Added LLM support for form documents (e.g. GewA1)

• Support for with value extraction with bounding box

• Updated stamp extractor that allows user to detect and extract stamps from documents

• Enabled superTask for the verification task
• Table Editing: added ability to expand the view across the entire width and / or height of the screen via expand button and orientation switch

• Fixed a bug when changing TTL settings

• Replaced default color by specific instance color for selected bounding box
• Implemented pagination for thumbnail renderer for large amount of data points

• Enhanced support for extracting fields from tables via LLM

• Fix broken matching logic for Field Database lookup