GCRIPT - FAQ

A: GCRIPT (Gen-AI based Construction, Readability Improvement, and Punctuation Tool) is an AI-powered tool that automatically improves Microsoft Word documents (.docx) by correcting grammar, enhancing readability, and refining sentence structure.

Key Differentiators:

• Advanced Formatting Preservation: Maintains document structure, footnotes, hyperlinks, tables, and most formatting elements
• Intelligent Text Anonymization: User-specified sensitive text is anonymized at the paragraph level, while numbers are automatically and intelligently anonymized such that context and sentence meaning are preserved
• Comprehensive Document Support: Handles complex documents with tables, merged cells, images, and various formatting elements
• Secure Processing: Sensitive information never reaches the AI models in their original form

A: GCRIPT supports Microsoft Word documents (.docx files up to 16MB) including:

• Business reports and proposals
• Technical documentation with complex formatting
• Academic papers with footnote references
• Documents with tables, merged cells, and embedded images
• Files with hyperlinks and cross-references

A: Follow these steps:

1. Register: Select a username and a password
2. Access: Log into the GCRIPT web interface using your credentials
3. Upload: Select your word document (.docx, up to 16MB)
4. Configure: Choose processing options and add any sensitive terms to protect
5. Select Provider: Choose your preferred AI provider (Bosch LLM Farm recommended)
6. Process: Initiate processing and monitor progress
7. Download: Retrieve your improved document upon completion

A: GCRIPT offers three processing levels:

• Fast Processing: Skips paragraphs with 3 words or fewer. Our experience indicates that LLMs typically do not add significant value to very short paragraphs, allowing users to save processing time and reduce token usage.

• Balanced Processing: Processes paragraphs with more than 1 word. Provides more thorough improvement while maintaining reasonable processing efficiency.

• Complete Processing: Processes all paragraphs regardless of length. Offers maximum coverage for documents requiring comprehensive enhancement, though this increases processing time and token consumption.

A: Chunked Upload is designed to overcome limitations imposed by certain proxy configurations that may prevent standard file uploads. This feature splits your document into smaller segments for transmission, then reassembles them on the server. Enable this option if you experience upload failures with standard file transfer.

A: The Batching toggle allows GCRIPT to group paragraphs and table cells into batches for processing, which improves efficiency and processing speed. When enabled, multiple text elements are processed together rather than individually, reducing overall processing time.

A: GCRIPT employs sophisticated formatting preservation techniques:

• Style-Based Formatting: Formatting is best preserved when documents use defined styles rather than direct formatting
• Paragraph-Level Preservation: Formatting is maintained at the paragraph level, as sentence-level changes may result in text modifications that prevent exact formatting mapping
• Hierarchical Formatting Considerations: In cases where formatting appears altered, this may be due to Word's hierarchical formatting system where local overrides supersede style definitions

A: GCRIPT supports comprehensive document structures:

Fully Supported:

• Tables, including those with merged cells and embedded images
• Images within documents
• Hyperlinks and footnote references
• Standard formatting elements

Limitations:

• Embedded objects, shapes and text within shapes are not currently supported
• EndNote and intra-document references/fields are not supported and require manual rectification/ re-insertion

A: GCRIPT processes references (footnotes and hyperlinks) on a best-effort basis:

• Hyperlinks and Footnotes: Where possible, automatically reinserted after the text has been improved by a LLM
• Failed Reinsertion: When automatic reinsertion fails due to text changes and inability to find context, the tool flags this with red text at the end of the relevant paragraph
• User Action Required: Users can manually reinsert flagged references and remove the warning text
• Future Development: EndNote and intra-document references are planned for future releases

A: Use Microsoft Word's Compare Documents feature:

1. Download and save the document after processing
2. Open Word and go to Review > Compare > Compare Documents
3. Select your original document and the GCRIPT-processed version. You may want to unselect the 'formatting' option to reduce clutter.
4. Word will highlight all changes, additions, and modifications
5. This allows you to review exactly what improvements GCRIPT made

A: GCRIPT implements multiple security layers:

• User-Defined Protection: Add sensitive terms/ phrases to the "Protected Words" section before processing. These are anonymized before text is sent to the LLM for improvement
• Automatic Number Anonymization: Financial amounts, dates, IP addresses, and other numbers are intelligently anonymized while preserving context
• Paragraph-Level Anonymization: All sensitive information (text and numbers) is anonymized at the paragraph level to reduce the probability of correlating information
• Restoration Process: All original sensitive information is restored before the final document is presented for download
• Note - sensitive term replacement is case-insensitive (google will match Google, gooGle or other combinations) but replacement back is as per the original case (so if you had gooGle in your orignal document, post improvement of the text by the LLM, you should see gooGle in your downloaded document)

A: Yes. GCRIPT is a PoC and we are gathering data to improve the application. Accordingly access to uploaded and improved documents is restricted to one designated individual solely for performance evaluation and feedback validation. Temporary files are automatically deleted after processing is completed by the application.

A: Input Token usage includes detailed instructions, which are sent to the LLM with each paragraph or batch of paragraphs. The development team is actively working to minimize token consumption to improve processing speed and optimize costs.

A: The GCRIPT team is continuously working to make processing more efficient. Recent optimizations include:

• Smarter Batching: We've improved how GCRIPT groups your paragraphs before sending them to the AI. By more accurately predicting the token count, we can safely fit more of your text into each request, reducing the total number of calls made.
• More Efficient Structure: We've reduced the overhead required to structure the data sent to the AI. This means more of your token budget is spent on improving your content, not on formatting instructions.

These changes result in faster processing times and lower overall token consumption, especially for large documents.

A: No cost is reflected when Bosch LLM Farm is chosen as a provider as we are extending the benefit that Bosch LLM Farm has provided for this PoC. Other AI providers charge at actual rates and these are refleced after document processing. However, this is not passed on to users as these are currently being borne by the development team personally. Further on account of data privacy, you are encouraged to utilize the models offered through the Bosch LLM Farm.

A: Processing duration varies based on document complexity and size:

• Small documents (1-5 pages): 1 -2 minutes
• Medium documents (5-20 pages): 2 - 5 minutes
• Large documents (20+ pages): 5+ minutes
• Processing time also depends on the document structure (number of paragraphs, tables and their arrangement). AI provider response rates and system loads also contribute.

A: Try these solutions in order:

1. Verify file format: Ensure your file is .docx format
2. Check file size: Confirm the file is under 16MB
3. Enable Chunked Upload: Use this option for proxy-related upload issues
4. Try a different browser: Chrome, Firefox, or Edge typically provide optimal performance

A: Common resolution steps:

1. Switch AI providers: Try a different provider if one fails. Note the system also has an automatic fallback capability, where it will switch models and then providers if it detects failure
2. Reduce document complexity: Process a smaller or simpler document first
3. Verify connection stability: Ensure reliable internet connectivity
4. Contact support: Email dkx2kor@bosch.com if issues persist

A: Yes. Use the "Cancel Processing" button to terminate processing at any time. You can then modify settings and restart the process.

A: GCRIPT currently supports:

• File Formats: Microsoft Word .docx files (we plan to look at other file formats too later)
• File Size Limit: Up to 16MB per document
• Compatibility: Files created in Word 2007 or newer

A: Current limitations include:

• Shapes and text within shapes are not yet supported
• Embedded objects are not yet supported
• EndNote and intra-document field references require manual verification/ re-insertion
• • Section Breaks within Paragraphs: If a section break occurs in the middle of a paragraph, the paragraph may be processed twice. While this does not harm the document, it can lead to slightly increased processing time. We recommend placing section breaks between paragraphs for optimal performance.
• Very complex nested formatting may require review. We recommend using the 'Compare Document' feature of word to easily identify and review and accept/ reject changes

A: We actively encourage feedback on all aspects of GCRIPT. We are actively working to make a feedback section available for submission of your feedback. In the meantime please send us your feedback through email - dkx2kor@bosch.com or sus1pu@bosch.com with subject line "GCRIPT Feedback: {Subject/ Topic of Feedback}"

We look forward to -

• Criticism (especially)
• Anything that doesnt work for you
• Performance issues or bugs
• Feature requests and suggestions
• User interface improvements
• Usability concerns
• Any other observations or recommendations

A: For technical support or detailed assistance:

• Primary Contact: dkx2kor@bosch.com
• Include: Specific error messages, document types, and detailed descriptions of issues encountered