Release Notes

Semantic Skills Enhancements

The following improvements are part of the release to streamline the creation, testing, and migration of Semantic Skills:

• Save Conflicting Skills for Review
  Developers can now force-save skills with conflicts in a "Pending for Review" state. This allows saving draft versions of skills even when conflicts exist, enabling easier reviews and adjustments before finalization.

• Skill Identification Accuracy Testing
  You can now generate test phrases for a skill and validate how accurately the skill is identified based on its Classification Details. This helps improve precision during skill training and QA. For more information, refer to Test Skill Identification Accuracy.

• Training Phrase Visibility for Migrated Skills
  Developers and administrators can now view the training phrases used to generate Classification Details for migrated skills. This ensures that all important user phrases are retained during the migration process.

Improvements to Luma AI Services

The following improvements to the Luma AI services are part of this release:

• LLM Resiliency & Failover Support
  Luma now supports automatic fallback to a secondary LLM in the event of outages or delays with the primary LLM model. This reduces reliance on a single provider and ensures seamless conversational experiences for users.

• Asynchronous Intent Categorization
  We've optimized the Intent Categorization Prompt with asynchronous processing, resulting in faster and more efficient response handling by LLM.

Other Enhancements

Here are the other functionalities that are available as part of this release:

• Toggle Default Knowledge Search: You can now enable or disable the default Knowledge Search triggered at the start of a conversation. This setting can be configured via Tenant Settings, offering more control over Luma’s initial behavior and improving initial response time.

• An animated status indicator is now available in the Test Widget and Web Widget, helping users stay informed and engaged while Luma processes their requests.

Intent-Based Skill Classification

Luma now supports Semantic skills. These are Intent-Based Skills that use AI-powered classification to define how a skill is recognized and invoked. These skills are built using structured intent parameters called Classification Details. These components are generated through a guided skill creation workflow, where Luma prompts the user for information and automatically constructs the classification logic. This approach eliminates the need for training data, reduces effort, and enhances the consistency of skills. Additionally, you can now publish the skills individually or bulk publish the classifications.

NOTE: To migrate your tenant from Phrase-based skills to Intent-based Skills, get in touch with Seviceaide Support.

For more information, refer to  Intent-Based-Skills-(Semantic-Skills)

Conflicting skill identification 

Luma can now detect potential conflicts between newly created skills and existing ones. This AI-based conflict ensures a clean, unambiguous, and well-structured skill repository. Skills with overlapping or similar purposes introduce execution errors and increased support overhead. Luma can now highlight the Reason for the conflict and provides Suggested Actions that the skill developer can act upon.

For more information, refer to Creating a new skill → Conflict-Check

Semantic Skill Search

No need to train Luma to identify skills. Luma now uses the power of AI to determine the most appropriate skill to invoke. Rather than depending on a fixed list of phrases or keywords, these skills use Classification Details to identify skill’s intent based on natural language. This method is flexible and adaptive to variations in how users phrase their requests, making it ideal for scaling Luma functionality in environments with dynamic or complex needs.

For more information, refer to Invoking a Skill

Natural Language Entity Recognition (NER)

Skip manually tagging attributes in phrases and let Luma automatically identify and populate the required information from the conversation flow. Luma uses AI to automatically populate all the attributes that are part of the skill (as user prompts) and the predefined tenant-level Common Entities. This intelligent entity recognition significantly reduces the manual effort required during skill development, accelerates go-live time, and improves the user experience by allowing Luma to gather more context from every user interaction.

For more information, refer to Invoking a Skill → Entity-Extraction-and-Attribute-Handling and Common Entities.

Responsible AI: PII/PHI Protection

Keep your personal information safe with PII. Luma VA introduces PII/PHI detection:

• Automatic Detection and Masking of PII/PHI: Luma VA uses advanced algorithms to detect Personally Identifiable Information (PII) and Protected Health Information (PHI) in real-time user conversations. Sensitive data is automatically masked in conversation history and support agent chats, minimizing the risk of accidental exposure, and ensuring compliance with privacy regulations.

• Flexible PII Configurations: Beyond a comprehensive list of out-of-the-box (OOTB) PII fields, organizations can define custom PII settings to meet unique requirements.

For more information, refer to PII Compliance in Luma

Luma AI Services Update

The following improvements in Luma AI Services are part of this release:

• GPT-4o mini is now the default LLM model for Luma AI- Prompt Templates. The model is configured by default during provisioning.

• You can now create custom prompts for OOTB AI use cases such as intent identification and precise answer generation. This enables organizations to align AI-generated responses with their domain-specific needs and language.

For more information, refer to Customizing Prompt Mapping

Other enhancements

Following are the other enhancements that are part of this release:

• Luma now handles the change of data type for local attributes. The latest data type will override the previous one.

• Agent Chat Console now focuses only on the ongoing user conversation. The Support agent sees only the current conversation. This ensures they do not have access to unauthorized information.

Two-Factor Authentication

Introducing Two-Factor Authentication (2FA) to enhance the security of Luma Skill Builder. Users can use a supported authentication app (such as Google Authenticator, Twilio Authy, Microsoft Authenticator) to generate a 6-digit verification code and complete a second login step. As an Administrator, you can enforce 2FA for individual or all users, preventing the option to disable it. In case of an error or loss of the device, users can reset the 2FA and recover the verification code with the authentication app. NOTE: This update does not impact the Luma Mobile App login.

For more information, refer to Two-Factor Authentication.

Enhancement to Domain/Group-Based Permission

Following improvements in Domain/Group Based Permission are part of this release:

• Domain administrators can manage users within their assigned group. Once a group is associated with a Domain, only the domain administrator (administrator in the group) can add or delete users in the group, ensuring exclusive control over group management.

• Enhanced access control on conversation logs in Luma Bot Builder: Conversations linked to restricted domains (managed by a specific admin group) can now only be viewed by permitted admins within that group. Default admins and other unauthorized admins will not have access to these conversation logs.

For more information, refer to Group/Domain-Based Permission.

Semantic Knowledge Search

Introducing Semantic Knowledge Search.  This release enhances Knowledge Search in Luma by utilizing vector embedding and full-text search to improve the identification of relevant artifacts, integrating Generative AI for precise answer generation, and structuring the search result set to provide comprehensive and accurate responses. The key benefits of Semantic search are:

• Zero-Touch Curation: No manual keyword extraction or metadata fine-tuning.

• Accuracy Boost: Accuracy improved to over 94%.

• Enhanced User Experience: Natural and precise responses to every user query, leveraging Large Language Models (LLM).

• Multi-Source Answers: Generating answers from multiple knowledge articles.

• RAG Implementation: Answers are generated based on the knowledge base, ensuring reliability.

These updates streamline curation, improve accuracy, and enhance user interaction with semantic search.

Note: All new tenants will be provisioned with Semantic search by default. For existing tenants, get in touch with the Operations team to enable Semantic search. This would require migrating the artifact structure to support semantic search i.e. reprocessing of all the Artifacts available in the Tenant. This is a non-reversible process. Once migrated, cannot be reverted.

For more information, refer to Semantic Search.

Enhancements to Visual Designer

The following updates are done to support the improved Knowledge Search:

Search Knowledge Step: 

• Tenant-level configuration parameter “Return all matched articles during initial semantic search” to return matched articles along with precise answers and source articles.

• “Return all matched articles” to return matched articles during ad hoc Knowledge searches.

• Knowledge search responses now include the Precise answer, Source Articles used to generate Precise answer, and other Relevant Articles.

For more details refer to Search Knowledge

Display Precise Answer:

• Source articles can be presented in cards after the Precise answer. You can configure the step to display the source. This is configurable.

• Source articles will now open in the web browser if the corresponding article is an external document (has external redirection).

For more details refer to Display Precise Answer

Domain/Group-Based Permission

Enhances security by restricting access to pertinent conversation data to authorized users. You can now restrict access to conversation logs based on the domain or user’s intent, ensuring only authorized users from a specific department or group have access to conversation details.

As a tenant administrator, you can enable Group/Domain-based permission for your tenant. Once enabled, only the ‘Default admin’ can manage domains and corresponding admin groups/associated users. Domain admins can view conversations and unmatched phrases associated with their domain. A domain is identified based on the user’s intent or the skill executed in a conversation. Dashboard reports honor domain permissions, including only data from the logged-in user's domain.

For more information, refer to Group/Domain-Based Permission.

Other Enhancements

Following are the other changes that are part of the release:

• Web Widget: Avoid showing all data in the card text, instead show a view more/less similar to the card subtitle

• Avoid Initial Catalog Search if the tenant does not have Catalogs Click here for more details

• New "Skill Gap" KPI to identify the cases where skill training or new skills are required. Click here to know more

• Ability to configure global attributes to send in the message to web widget so that client can drive the user experience in the host application. For more details click here

• Ability to use global attributes to configure follow ups, channel notifications, support agent actions.

• An Import preview report before importing. You can unselect the items that may get over-written otherwise. For Details click here

AI Prompt Management

A centralized Prompt Management module that offers efficiency, customization, and enhanced user experiences, streamlining the Luma AI Services integration. As the skill developer, you can now create and manage prompt templates centrally for easy access, control, and collaboration across skills. Test prompts and tune configurations for optimized performance and productivity. Share easily with Export and Import of Prompt templates between environments.

For more information, refer to Prompt Management.

JavaScript Support

Script support for advanced customization capabilities. You can now add JavaScript to your skill to handle complex scenarios such as performing calculations, data manipulations, or other custom logic.

You can use JavaScript in your skill to:

1. Assign values to an attribute in the Set Attribute step and Integrations

2. Validate an attribute using Attribute validations

3. New Conversation Item: Script step to facilitate processing and setting context attributes using JavaScript

For more information, refer to Using Script.

Advanced Dashboard (Superset) - Beta

Advanced Dashboards to view visually rich data visualization, customizable dashboards, and integrated reporting capabilities. The new dashboard empowers users with actionable insights, facilitating data-driven decision-making and enhancing the overall performance and efficiency of bot interactions. You can transform KPIs into visually compelling charts and graphs for enhanced data interpretation and decision-making.

For more information, refer to Advanced Dashboard (Beta).

Visual Designer enhancements

Following are the visual designer enhancements that are part of this release:

• 'AI Tasks' Conversation is now renamed to 'AI Prompt Template'.

• Extending the scope of AI Prompt context variables. Values in the AI Prompt (AI Task) Context variables can now be accessed and referred to through the skill execution, similar to all other context variables.

WhatsApp channel enhancements

Improved card display on WhatsApp Business. Luma now uses the Interactive Message API to send cards, making Knowledge Articles appear as structured cards instead of unstructured text.

Agent console enhancements

Attachment support is available in the Agent chat console for analysts to send attachments to users in the chat.

For more information, refer to Agent Chat Console.

Other Enhancements

Following are other enhancements that are part of this release:

• Display up to 10 skills to the user if Luma encounters an ambiguity while looking for relevant skills.

• Configure WhatsApp-specific messages for Suggest Skills.

• A new utility to translate the skill content from one language to another.

• Viewing the object definition (skill, subflow, attribute, operation) from conversation events.

• Introducing the latest GPT4 & GPT-4-Turbo models in AI Prompts. You can now create prompts and use the latest GPT4 & GPT-4-Turbo models for optimized performance and accurate response.

• Ability to tune an Adhoc AI Prompt's deployment model and response parameters.

Support for reusable Subflows

Enhance your skill-building experience and streamline conversations with reusable subflows. Subflows are reusable 'Functions' within a skill, eliminating the need to branch to another skill. This promotes reusability by allowing subflows to be used across multiple skills, enhancing efficiency and consistency in your development process. The sub-flows can be easily created and managed under the Subflow section. Add the sub-flows to your skills using Conversation Step 'Execute sub-flow'. 

Customize system skill behavior

Enhance user experiences by personalizing System Skills based on your organization's unique requirements. System skills, like User Registration, Welcome, What can you do and Transfer to agent, are now customizable, allowing you to tailor them to your specific needs instead of triggering out-of-the-box functionality. You can configure a skill to be triggered when the System Skill is requested.

Business Domain Management

Enhance security and control while tailoring interactions to specific business needs. Classify customer business domains at the tenant level and configure conversation permissions based on domain status; A whitelisted Domain for unrestricted access, Blacklisted Domains for blocked domains, and Greylisted or Unclassified domains for conditional approval. 

For more information, refer to Business Domain Management

New Conversation Steps

The following Conversation Steps are available as part of this release:

• 'Categorize User Intent' using User Phrase to categorize the user's phrase

• AI NLQ Step to determine the Change gear ticket Query

• Execute Sub flow to execute sub-flows as part of skills

Support for New start up

New startup skills integrated with Generative AI and Changegear OOTB skills

Customized Startup Skills for standalone VA

Ability to enable custom startup behavior for standalone VA customers and configure skill presentation template. 

Support for Monthy Session-based licensing

The following updates are part of the release to support Monthly Session consumption:

• Luma now supports a new license type Sessions Per Month is now available for New tenants. The older Session Based licensing is not available for new tenant provisionings

• Auto-renew sessions every month until the end of contract.

• Ability to update Sessions in the packs of 500 by default, but can be manually updated. It can only be incremented, but not reduced below the consumption

• Monitor the usage and alert when the consumption reaches 75%, 85%, 95% and 100%. Continue to alert if the consumption is >100% until the end of the current month

• License Purchase History show the both manual purchase and auto-renewal activities

• Monthwise session consumption available on Usage Dashboard (and report) and Tenant Settings->Subscription page
  
  
  

Visual Designer Improvements

The following updates Visual Designer updates are part of this release:

• Ability to select one or more steps in the skill and create a Sub flow with the steps.

• Search Catalog step: Ability to exclude previously presented search results

• Search Skills step: Ability to exclude previously presented search results

• Search Knowledge step: Ability to exclude previously presented search results

Provisioning Luma Knowledge with Luma Virtual Agent 

VA & KM can now be  provisioned together via the Luma Super Admin portal. As a super admin, you can choose to provision Luma Knowledge, Superset and Automation at the time of Luma Va provisioning or later. The system would automatically provision the tenants and correlate. This also irradicates the need to execute User sync.

Note: New Luma Knowledge tenant must always be created via Luma VA Super admin screen. For correlating Existing tenant, please reach out to support team. 

Other Enhancements

Following are other enhancements that are part of this release:

• Support Agents can now see the User Conversation start and Elapsed time meter on the Agent Chat Console.

• Enhance the user experience by providing information on the source of the Knowledge. Luma now displaus the Artifact name and the source along with the Knowledge search results.

• We can now record the requests that are resolved using the information available over Internet. We can using Set KPI step to calculate KPI "Resolved with External Information" whenever a user request is resolved with information from Internet

• The default configuration for Browse Services system Skill is now Quick Reply.
  Ability to execute the skill presented in Suggest Skills in the context of the current conversation

Visual Designer Enhancements

Improving the usability of Visual Designer, the following updates are part of this release:

• Support for Collapse view: Easily navigate through your complex skills in the collapsed view. The feature enables the skill developer to manage design space on the canvas efficiently, ensuring that the design environment remains clean, organized, and readable.

• Enhanced search functionality within a skill. In addition to searching by step names, the system now allows the users to search through the objects used in each step.

• Ability to clear the local attributes during the conversation.

• Toggle the properties panel within the Visual Designer. This allows the users to see the properties panel or hide it, depending on their current tasks or preferences.

For more information, refer to Defining the skill with Visual Designer

Troubleshooting Logs for Follow-ups

Get better insights and control over your configured Follow-ups. Following are the updates that are part of this release:

• Dive deep into the Follow-up polling process with comprehensive troubleshooting logs available at each configuration level. Identify and address issues swiftly, ensuring a seamless follow-up experience.

• View Follow-up data polling troubleshooting logs and resolve any issues promptly.

• View the active follow-up data records in real-time.

• Show message extraction events in follow-up conversations

For more information , refer to Follow-up

Improved Skill search

Luma now uses SBERT (Sentence-BERT) in collaboration with RASA to provide a more robust and sophisticated intent identification system. This integration ensures more accurate skill matching. This is an optional feature and will be rolled out for only specific customers initially. 
To enable SBERT for your tenant, please reach out to the Support Team.

Note: Enabling the SBERT may change the results for the existing skills. You should test thoroughly and adjust the NLP Configurations as required. 

Other Improvements

Following are the other changes that are part of this release:

• Deprecating Classic Designer for new skill creation. However, there’s no change in view/edit of existing skills using the classic designer.

• Ability to send payloads for GET Rest calls

• Show suggestions from the last skill after the survey is completed

• Need support to multiple same query parameters in outbound REST calls

• To improve Luma performance, the troubleshooting event logs are now retained only for 7 days. The data is purged every day. There will be no impact on the Dashboard and Reporting.

Enhancement to GUI Skill Builder 

Bridging the gaps between the Visual Designer and the Classic Designer, the following updates in Visual Designer are done as part of this release:

1.  

   • Support for Attribute extraction from user phrase for the 'Date' type attribute. Luma now extracts the date provided by the user in the search phrase and assigns the value to the 'Date' type attribute encountered first during execution.

   • Support for Capture Feedback Step

   • Support for Capture Satisfaction Survey Step

   • Support for Triggering Automation Workflow 

   • You can now migrate your classic designer skills to Visual Designer skills, making it easier to manage and update.

For more information, refer to Building Conversation with Visual Designer.

Improvements to Visual Designer

The following new updates are part of this release:

• Support for Search Skill. You can now perform a custom search on your pre-configured skills from your visual designer skill and display the results during conversation

• Updated Prompt- Ability to configure global attributes to send in the prompt reply message to the web widget so that the client can drive the user experience in the host application

• Multiple Step deletion: You can now effortlessly select and delete one or more steps from a skill. This removes the inbound and outbound connections from the selected controls

• Support for keyboard shortcuts to Copy (ctrl+c) and Paste (ctrl+v) of one or more elements

• Support for Create Artifact to create a Knowledge Artifact in your Luma Knowledge tenant as part of your skill.

Integrating AI services

As part of continuing efforts to integrate Generative AI in Luma Virtual Agent, the following updates have been done in this release:

• Luma is now fully integrated with Luma AI services. Harnessing the power of advanced Prompt engineering, the Generative AI functionalities in Luma now uses AI services. The integration streamlines the metering and management processes for optimal utilization.

• Introduction of new conversation step "AI task" to execute an AI Task as part of a skill. Using the step, a skill developer can using OOTB Luma AI services or create an ad-hoc task. The system captures the usage for each customer for reporting purposes

• Ability for Tenant Admin to enable/disable AI Task in the conversation flow

For more information, refer to Generative AI in Luma

Centralized Generative AI Configurations

Centralized Generative AI Configurations for Luma Suite of Products. Empowers Superadmins to effortlessly configure and oversee the Generative AI functionalities for all Luma products, simplifying management and optimizing control through one consolidated platform. A Superadmin can enable/disable Generative AI and configure OpenAI credentials for a new or an existing tenant with the ability to enable/disable features

For more information, refer to Configure Generative AI Configurations.

Schedule Notification

You can now schedule notifications in Luma. The notification can be scheduled via the Bot builder interface by the Tenant administrator or can be incorporated in Skill using the REST APIs.  You can create one-time or recurring notifications

The schedule also supports dynamic information, recipient lists, or schedules using the notification API. The information can be used to generate Notifications.
View the scheduled and triggered notifications in History
You can also export and import notification configurations

Notifications on the WhatsApp channel can also be sent to unregistered users

VA-KM unification

In the process of unifying Luma Knowledge and Luma Virtual Agent, the following updates are done as part of this release:

• Luma Knowledge is now available as a workspace within Luma Bot Builder with implicit login. You can now launch your Luma Knowledge Curation portal directly from Luma admin portal, without the need to log in again.

• Schema changes to support VA and KM Unification.

Other enhancements

Following are the other enhancements that are part of this release:

• New Conversation events for AI TAsk and Create Artifact steps

• Updates to troubleshoot logs to show more details in case of error with Set/Transform attribute. You can now see the attribute that causes the error in the logs along with already filled attribute values.

• System now captures Luma and 3rd party integration response timings, which can be used for reporting and troubleshooting

• Search Skill and Display skill controls are now available for VA-only customers to support any fallback scenarios.

• When a quick reply/card button is clicked, you will now always see the button label instead of payload/value 

Enhancement to GUI Skill Builder 

Bridging the gaps between the Visual Designer and the Classic Designer, the following updates in Visual Designer are done as part of this release:

1.  

   • Support for Search and Display Knowledge Articles- You can now perform a custom search on your Knowledge base and display the results as part of your visual designer skill.

   • Support for Search and Display Catalog to perform a custom search on your Catalogs and display the results as part of your visual designer skill.

   • Support for Display Skill found as required in your skill.

   • Support for KPIs Step in Visual Designer to update Luma VA and Luma Knowledge KPIs through the visual designer skills.

   • Support for Suggest Skills to configure the skill suggestions to be presented to the user as part of the skill. You can now configure a list of skills or use the skill suggestions categories use to present skill suggestions to the end user.

   • Support for Image type in Messages and Prompt

   • Update to Trigger Skill Step: Trigger a Skill dynamically i.e. using a variable that holds the skill identifier.