• 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.

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

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

• 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

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

‍

• New Email Attachment Scanning UI

• 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

• 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.

• 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

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

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

• Updated document types: Invoices, contracts

• 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"

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

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

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

• Enhanced stamp detection

• 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

• Several minor usability improvements

• 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

• Bug fix regarding token usage estimation

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

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

• Introduced new document type: crash report

• 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

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

• Enhanced hints for prompt input

• Fixed a bug regarding ground truth examples in form documents

• Fixed a bug regarding multi-chunk extractions

• Fixed a bug that caused missing extractions in long documents

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

• 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

• Fixed a bug regarding optional annotations

• General performance improvement for AI predictions

• Improved bounding box support for LLM extraction

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

• 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.

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

• 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

• 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

• Performance improvement for displaying jobs with big payloads

• Fixed LLM token calculation

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

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

• New LLM options for fields

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

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

• 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

• Fixed a bug in LLM prompt examples

• Redaction now relies on new AI framework

• 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

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

• Table Editing: Support for vertical headers

• Table Editing: Multi-page tables highlighting support

• 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

• Fixed a bug when changing TTL settings

• 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

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

• Fix broken matching logic for Field Database lookup

• Enhanced support for extracting fields from tables via LLM

• Added ability for users to define a job time to live (TTL) on project level

• Improved composition of LLM input prompts

• Fixed a bug so that only the necessary information is redacted on a word level
• Fixed a bug that led to inconsistencies in the confidence scores
• Fixed a bug in the Fuzzy Matcher that led to decreased quality of the output

• Fixed a metric computation issue

• Fix a bug that shifted the link between annotations in document and the value panel
• Fixed a bug that led to cut off the last item on the value panel

• Support for filename column in Work Queue

• Fixed a bug that in some case led to incorrectly validate LLM response
• Fixed an input prompt issue

• LLM support for Custom Table Headers

• Metric computation enhancements

• Support for coarse-grained confidence scores

• Fixed an issue with resolving data URLs for calculated field datasets.

• Fixed a bug in the fuzzy matcher
• Fixed multiple bugs in training and prediction pipieline

• Added support for multi-page tables: Merge tables on different pages into one table output

• Fixed a bug that didn't add scroll bars in oversize text inputs
• Adjusted Worker modal for super tasks to render updated schema properly
• Fixed editing confirmation dialog for new tables

• Document.Extract widget: fixed a bug related to review icons for tables.
• Document.Exrract widget: fixed a bug that led to inconsistencies when persisting matched table headers.

• Document.Extraction widget: fixed unnecessary confirmation dialog during annotation creation
• Increased font size in annotation popup
• Added scrolling to selected field in annotation fields pane

• Document.Extract widget: fixed selection of multiple annotation on Mac.
• Document.Extract widget: fixed a bug so the the modified table cell on the table panel is always persisted.

• Document.Extract widget: ask the user to confirm selection of another annotation if the currently selected table has been modified.
• Document.Extract widget: change algorithm of table scaling. As many rows/columns as possible keep their position.

• Fixed a bug in Job Design preventing the user to select fields to redact.
• Adapt types for invoices to comply with model returns. In some cases this bug was leading to invoices without any extraction due to a schema validation error.

• Add fuzzy matcher after the model call for invoices, enabling fields to be found by key value pair matching.

• Allowed and blocked list support for entity extraction
• Redaction support for single-page and multi-page tables.

• Batch support in S3 integration

• Fixed a bug in the PDF format check
• Fixed appearance of "Add" button in HITL task

• Added support of multi-page table rendering
• Allows the users to split and merge tables on different pages

• Improved OCR side wagon

• Fixed a bug that led to "Can't show image" message after Edit Output

• Added new model option for invoice processing

• Allow users to login with MFA enabled for all users

• Fixed a bug in displaying tables when the row / column span is larger than 2.

• Rotate images and documents to their correct orientation if page rotation is detected

• Fixed a bug that prevented job submissions for instructor and member collaborator roles

• Fixed a bug that led to calculated field not working with redaction option activated

• Added ability to define custom table headers for consistent table column extraction

• Added a redact post-processing step with field level redaction settings to blacken or blur parts of documents
• Added ability to rename file names for CSV download based on extracted field values

• Updated error handling

• Database lookup: allows users to link a dataset to a particular field and returns best matching candidates.

• Edit Job: fixed a bug that led to lost modifications of cell content

• Edit Job: Enhanced table content menu

• Fixed a bug that led to crash the extraction widget when adjusting table dimensions

• General availability to a call to detect and correct document rotation.
• Customize CSV download: The user can include or exclude annotations from CSV files.The user can flatten the tables in CSV files. User can exclude CSV headers, set CSV field delimeters, and quote values

• Customizable CSV download: user can exclude header, can set the field delimiter and can set to quote the values

• Fixed a bug in table matching that was resulting in invalid schema

• Fixed a bug in Document.Extract widget that led to incorrectly convert values after editing field.

• Enhanced editing capabilities on Tables via the context menu
• Better support for merging / splitting table cells

• Fixed a bug in the combiner that was at times deleting polygons twice

• Combine key value pairs from different AI models as an input for the fuzzy matcher

• A new query param is added to the `/v1/apps/{app_id}/job_responses` endpoint - called ids. If you pass job ids into that list only those ids will be in the data. Enables users to filter the results by ids.

• Fix a bug that led to orientation issues due to existing metadata present in images

• Combine different field detection sources in one single output to increase accuracy and recall

• Improved AI model response time

• Fixed a bug in the Document.Extract widget that validates the output before submitting the results. All the errors are highlighted in the widget now.
• Fixed a bug that led to incorrect number of lines in created tables
• Fixed a bug that wrong date format
• Multiple minor bug fixes

• Allow empty strings as value for the prediction in General Document Processor.

• Enhanced key-value pair extraction in the General Document Processor

• Support for signature detection and extraction in the General Document Processor

• Fixed a bug so that the callback reference submits latest changes

• Fitted document to the width on initial loading, by clicking fit button.
• Enhanced vertical scrolling of a fitted document
• Enhanced support for merged table cells (colspan/rowspan)

• Fixed a bug that led to dropdown options not rendering nicely
• Fix a bug that degraded experience editing in Document.Redact
• Fixed multiple bugs in Table Editing widget
• Fixed a bug that led to an incostency when no document type was selected in design phase in onboarding
• Fixed annotation field popup positioning

• Usability improvement by adding shortcut to Edit Output modal

• Enhanced usability scrolling behavior of the key value pair panel
• Improved usability of the date input component

• Document validation checks if PDF have more than 0 pages

• Enable direct file submission for files below 4 MB

• Fixed a bug that led to not catching all request exception‚ while trying to load document

• Enhanced error reporting when document input is corrupted

• Fixed a bug that led to the field details modal would only disappear on toggle
• Fixed a bugs related to moving annotations to another field
• Fixed a bug that lead to wrong order of custom fields in the Job Review

• Document.Extract: model confidence shown next to annotations

• Document.Extract: Fix a bug that was not returning strings as predictions
• Document.Extract: Fixed a bug that led to wrong data URI output in AI tasks

• Document.Extract: Removes duplicates from GewA1 doc type

• Document.Extract: The General Document Processor now supports extracting data from GewA1 (German: Gewerbeanmeldung) document type

‍

• ChoicesField correctly shows validation errors.
• Fixed several bugs related to key-value pairs in the Document.Extract widget.

• Refactoring annotation modal styles, added ability to delete selected bounding box.