Added regex functionality to deny lists. Corrected tesseract to word level parsing. Improved review search regex capabilities. Updated documentation

Allow for tesseract to run OCR in line-level mode and then query LLM with line-level data. Added option for running as MCP server, added api for multi-word text search

Fixed minor bugs related to Textract API calls, pyproject format. Removed print statements and fixed some future concat deprecation issues

formatter and linter applied

Added suffix to textract output files according to tasks included (e.g. signature analysis). Improved reporting when Textract client doesn't exist. Fixed display for cost and time taken. Changes to config variables to allow exclusion of PaddleOCR from display

Improved paddle and hybrid OCR analysis across all options. Tried to revise requirements for spaces

Allowed for load Paddle at startup. Updated requirements for torch compatability

formatter and linter applied

Updated word segmenter code

Added text rotation capability

Optimised VLM model choice and prompting/parameters

Added hybrid paddle + vlm option. Optimised word segmenters for single words. Optimised package installation in pyproject.toml

Added upgraded line to word parsing algorithm. Added dependencies and framework for Huggingface spaces deployment with ZeroGPU

Improved new requirements. Improved visual OCR outputs and word-level Paddle outputs and general bounding box positioning

Initial commit for VLM support. Created visualisations for OCR output. Corrected log_file_output_paths reference.

Corrected environment variable file references for log files and spacy/paddle folders for lambda_entrypoint

Formatter and linter check

Modified Dockerfile and entrypoint to switch user at runtime. Updated output folder file creation for custom_image_anlyser_engine and find_duplicate_pages.py

Updated formatting and linting check

Updated file save sections in custom_image_analyser_engine.py with secure file writes

Updated formatting and linting

Improved PaddleOCR implementation (greater accuracy, now can save outputs with config setting). Updated Dockerfile entrypoint for Lambda to hopefully avoid permissions issues

Updated paddleocr implementation to have a menu option on the GUI with a config value change. Minor package updates, favicon update, and update to Dockerfile to allow for Lambda function execution.

Added the possibility of saving initial redacted pdfs with redaction comments directly attached. Fix for missing Textract pages. Better Textract forms element extraction and save.

OCR outputs now return confidence values

General code changes and reformatting to address code vulnerabilities highlighted by codeQL scan, and black/ruff repplied to code. Fixes/optimisation of Github Actions

Fixed on deprecated Github workflow functions. Applied linter and formatter to code throughout. Added tests for GUI load.

Added example data files. Greatly revised CLI redaction for redaction, deduplication, and AWS Textract batch calls. Various minor fixes and package updates.

Corrected some multiple xlsx/docx file redaction issues. package updates.

Corrected an issue with finding valid language entities for AWS comprehend redaction

Updated command line redaction script with more options

Improved language support functions and reporting

Added support for other languages. Improved DynamoDB download

OCR results now return line numbers consistently. Made redaction search more resilient to punctuation at end of terms.

Added PaddleOCR support

Can now redact terms using a new redact search tab on the Review Redactions tab. Various minor improvements

Now local OCR outputs can be saved to file and reloaded to save preparation time. Bug fixing in logs and tabular data redaction. Update to documentation

Improved logging format a little. Now possible to save logs to DynamoDB

More config options. Fixed some bugs with removing elements from review page and Adobe export. Some UI rearrangements

Integrated AWS Comprehend and fuzzy matching functions with tabular data redaction.

Allowed for Textract and Comprehend API calls through AWS keys. File preparation function incorporated into main redaction function to avoid needing user to 'check in' during redaction process

Fuzzy match implementation for deny list. Added option to merge multiple review files. Review files from redaction step should now include text.

Ensured the text ocr outputs have no line breaks at end. Multi-line custom text searches now possible. Files for review sent from redact button. Fixed image redaction (not review yet). Can get user pool details from headers. Gradio update.

App should now resize images that are too large before sending to Textract. Textract now more robust to failure. Improved reliability of json conversion to review dataframe

Greatly improved regex for direct matching with custom entities

Started adding in support for custom deny list. Fixed textract call issue. Removed multithreading for now as it mixes up pages

Only shows AWS options when AWS functions enabled. Can now upload previous review files to continue review later. Some review debugging.

Comprehend now uses custom spacy recognisers on top of defaults. Added zoom functionality to annotator. Fixed some pdf mediabox issues and redacted image output issues.

Consolidated AWS Comprehend redaction calls to reduce total number

When on AWS, now loads in a default allow_list to exclude common words from redaction. Improved checks on AWS Comprehend calls.