- Improved queuing algorithm for a smoother bulk parsing experience
- Implemented stricter limits on email uploads
- Made it easier to discover the change collection menu
- Rebrand for consistency with new frontend website
- Added new validation tool config to disable currency formatting of decimal values
- Added requirement for valid currencyCode field+value for Invoices, Receipts and Credit Notes
- Significant overhaul of the new (Beta) tables interface to improve stability and user experience
- Added setting to workspace settings allowing users to control the sensitivity of the document splitting model
- Enable users to zoom in on documents in the splitting interface by clicking on the thumbnail
- Allow users to see when non-base extractors were last trained
- Allow users to control if a collection trains its (non-base) extractor
- Add field and datapoint to the annotation deleted front-end event.
- Fixed issue with new table fields not displaying correctly in the field config editor
- Stabilization for search scores.
- Fixed custom receipt table parsing
- Improve document upload experience in free tool
- Allow users with customised models to parse at a higher rate
- Significant speed improvements for parsing with OpenAI
- Fixes to bullhorn integration authentication flow
- Small fixes to frontend tool UI
- Fix for bullhorn authentication flow
- When a model is fine-tuned, we now reparse up to 100 unvalidated documents to take advantage of the improved model
- Fine tuned improved document classifier
- Improved stability of magic tables interface for non-invoice document types
- Better error handling
- Improved table detection
- Backend bug fixes relating to annotation display for rotated documents
- Improved GPT error handling
- Fixed issue where identical annotations on different pages were not being rendered
- Automatically close field editor when drawing new annotations
- Improved performance for detecting large text spans with openAI
- Allow magic table fill to be turned off at a workspace level by administrators
- Emphasise new field button when collection has no fields
- Fields that can be selected multiple times are no longer greyed out after a single selection
- Fixed an issue preventing reordering fields
- Make it easier to discover the configure fields button
- Fixes to usage tracking
- Improvements to magic tables
- Improvements to custom field predictions
- Improved performance in annotation handling (stage 1)
- Various backend bug fixes
- Allow users to add data points that don't have bounding boxes on the page
- Improved document sorting algorithm
- Improved construction of nested annotations
- More graceful handling of errors around currencies
- Other backend bug fixes
- Fix bug in scenario where document does not have a workspace specified
- Better handling of huge images during OCR
- Improved parsing of dates from text
- Removed deprecated db tables
- Only use documents within selected collection when parsing with GPT-4
- Improved grouping of annotations when using group annotation types.
- When words contain a forward slash, split them into two words
- Backend bug fixes
- Handle error with unidentified images
- Fix null handling in resume sections
- Improve handling for scenario when two tables are shown side by side
- Parse quantities annotated as "free" as 0
- Removed legacy Azure OCR service
- Fixed issue where renaming table fields could remove nested fields
- Show in the validation tool if a document has been split
- Show in the document list if a document has been split
- Magic tables bug fix that was causing row items not to be visible
- Additional setup for automatic document sorting
- Improved rate limiting for custom extractors
- Various backend bug fixes
- Support for nested field configurations
- GPT parsing for picklists, nested fields, multiple fields, and rectless fields.
- Improved handling for ingestion of documents via email
- Improved email sender handling
- Fix tooltip persisting after hovering field in sidebar
- Improve parsing of floats
- Various backend bug fixes
- Better signup flow for bullhorn integration
- Improving release process