Advanced Agent QA, Webchat Accessibility, and More with Cognigy.AI v4.84

Authors name: Nhu Ho September 13, 2024

The latest release of Cognigy.AI introduces significant improvements in Agent QA, Knowledge AI, and Webchat accessibility.

Optimize QA Processes with Advanced Debug Mode

Cognigy’s new Debug Mode replaces our previous Expert Mode and provides an advanced, intuitive tool for QA and troubleshooting AI agents during development. Seamlessly integrated within the Interaction Panel, it offers quick, granular, and real-time visibility into critical elements of Flow execution.

With Debug Mode, leverage a more robust and comprehensive set of diagnostic insights, including:

Triggered Intents and Scores: Instantly see which user intents have been recognized during a conversation and validate their accuracy.
Matched Slots and Lexicons: Monitor entities captured within each interaction, ensuring precise data extraction.
State Changes: Track how user sessions evolve, providing transparency into how states are managed and updated throughout the interaction.
Flow Transitions: Review how the system navigates between different Flows, helping to identify and resolve logic issues in complex conversation structures.

Debug Mode setting is enabled by default. Additionally, Debug Messages can be triggered explicitly using the Debug Message Node or through debug actions in the Code Node.

-thumb.jpeg?length=1920)

Improved Search Filtering and Expanded Model Support in Knowledge AI

Following Source Tag editability, we have introduced another major improvement in Knowledge AI search filtering capabilities. With the latest release, you can choose between AND or OR operators in Source Tags.

This feature gives developers greater flexibility to fine-tune how search results are filtered and retrieved to ensure precise answers:

AND Operator: When selected, only results that match all specified Source Tags will be returned, making it ideal for narrowing down specific information.
OR Operator: Results will be fetched if they match any of the specified tags, offering broader, more inclusive search outputs when flexibility is needed.

On top of this, you can now leverage text-embedding-3-large and text-embedding-3-small models from OpenAI and Azure OpenAI for document ingestion and processing in Knowledge AI.

Enhanced Webchat Accessibility

To deliver a more inclusive experience for all users, we have introduced new features to make multimedia content more accessible for Webchat v3. These are particularly relevant for users with hearing impairments or those relying on alternative formats for information.

Video Captions URL Field: For the Say, Question, and Optional Question nodes, a new field has been added to provide captions when using video content. This ensures that video files in Webchat interactions are supported with captions, making it easier for users to follow the conversation without relying solely on audio.
Downloadable Audio and Video Transcripts: Ability to download text transcripts of both audio and video content. This functionality makes multimedia interactions available in text form, improving accessibility for users who prefer or require text-based alternatives.

Other Improvements

Cognigy.AI

Marked the GPT-3.5 family models as legacy and removed the deprecated text-davinci model

Cognigy Voice Gateway

Added support for RFC 3261 for REGISTER: The `