- Version: 2.51
- Release date: 9th July 2024
New Features and Enhancements
Document Search Skill Enhancements
The Document Search Skill enhancements, which were started in the v2.49 release, continue with the addition of LLM skill updates, improved search options including different search functionality, and new flow actions. This has all culminated in an improvement of search results, which in turn improves user satisfaction as the answers are conversational and more accurate to the questions being asked; new search options include:
- Standard Search - Used by the Query Documents flow action. Works well for traditional full text search.
- Vector Search - Used by the Query Document Chunks flow action. Enhances the search by utilising machine learning to improve its understanding of the question being asked along with the information within the documents uploaded.
- AI Search - Used by the Query Documents using AI flow action. This is conversational document searching, not only allowing a user to ask a question, but allowing them to ask follow-up questions from the content returned. The search attempts to provide a textual answer to the question, not just a link to a document, replicating how two people would have a conversation.
The best advice we can give when it comes to finding the best search method for your flow and content is to try it out. This is now made easier with by a new test feature within the Document Search skill where you can ask questions of your documents content using all the above search options.
New Flow Actions
- Query Document Chunks - Utilising machine learning and searching over each document in small sections, this search has an improved understanding of both the question and documents stored, allowing it to return more accurate results with part of the document's content along with a link to the file where the content was found.
- Query Documents using AI - Returns relevant contextual information from the documents stored that answers the question asked. The flow action attempts to answer the question in a conversational way, rather than simply providing a copy of the document's content, with the added benefit of the user being able to ask follow-up questions, just like we would in real life, with the goal of getting the perfect answer to their question.
- LLM Skill Document Search - Returns relevant contextual information from the documents stored by utilising an LLM configuration within the assistant's LLM skill. Like the above flow action, the user can ask follow-up questions, just like we would in real life, with the goal of getting the perfect answer to their question.
Improved Expert Assist Formatting for Answers
Expert Assist answers now have the same formatting options as QnA and allow bold, italic, links, bullet and numbered lists. Not only does this make Expert Assist answers more rich and visually appealing, but it also makes the promotion of Expert Assist question-and-answers to QnA much easier without the need to do any additional formatting. Formatting can be added to answers by simply clicking the required formatting option within the editor.
Expert Assist Workspace Improvements
Following customer feedback received, the following updates have been made to the Expert Assist Workspace page, improving the visibility of questions that are important to our users by updating the default sort order, and providing more one-click filter and sort functionality, including additional visual cues, for:
- Questions you have asked.
- Questions containing your preferred tags.
- Questions you are following.
Minor Updates / Bug Fixes
- Ability to search Conversation History by user.
- Removal of the deprecated Get Transcript flow action.
- Update AI model to a more recent version.
- Resolve layout issues when the Data Store contains long uninterrupted values.
- Make sure the copy to clipboard icon always uses the most recent version of the Embed Webchat script after changes are made.
- Update help pop-ups so that they use consistent padding no matter where they appear within Tenjin.
- Update the document count parameter within the assistant details page so that it always includes the correct number of documents, ignoring any filters set within the document skill.
Share this article
Comments
0 comments
Article is closed for comments.