Version 3.1.0 🎉

Welcome to Version 3.1.0! We're excited to introduce new advanced classifiers for NLU, handy Webchat customizations, and dataset enhancements to elevate your bot-building experience. Dive in to discover all the goodies we've prepared for you!

🗓 Release Dates

• Smartly.AI: 25th
• Virtual Agent: 26th

🤖 Advanced NLU Classifiers

Important: All the following NLU enhancements are available only for users of the SNP 1.06 pipeline.

Introducing XG Boost for Improved Classification

We're thrilled to introduce XG Boost as a new classifier option in our NLU system. Renowned for its optimized gradient boosting capabilities, XG Boost provides both high prediction accuracy and speed. Its advanced features like parallelization, regularization, and tree pruning further bolster your bot's performance in classifying intents and entities.

Fallback NLU: Your Safety Net for Enhanced Intent Recognition with GPT 3.5

When conventional NLU techniques don't quite hit the mark, our Fallback NLU feature comes to the rescue. Built on a cutting-edge Language Model using few-shot learning, this feature seamlessly takes over to refine intent detection and provide more accurate outcomes, ensuring a smoother user experience even when the main NLU system faces challenges.

Curious? Contact us to activate this feature.

FlauBERT: A Specialized Vectorizer for French Bots

For those creating bots in French, meet FlauBERT—a new vectorizer trained on an expansive and diverse French corpus. Developed using the state-of-the-art Jean Zay supercomputer by CNRS (French National Centre for Scientific Research), FlauBERT aims to improve your NLU pipeline's effectiveness. A video presentation is available for deeper insights into its capabilities.

FlauBERT significantly outperforms traditional TF-IDF vectorizers by offering a more robust text representation, which translates into better understanding and classification of user inputs.

📘

Coming Soon: More BERT Vectorizers!

Stay tuned as we continue to develop language-specific, transformer-based vectorizers for even more customization and effectiveness.

💬 Webchat Enhancements and Customizations

New Features:

• Customizable Input Types: Enforce specific input formats like tel, number, etc., for a more streamlined user experience. To activate, simply include the code output.payload.webchat_type_input = 'number';.

Improvements:

• Dynamic Text Direction in Webchat: The webchat layout now intelligently switches between Left-to-Right and Right-to-Left orientations based on the active language.
• Localized Surveys: Surveys are now multi-lingual, adapting to the user's current language setting for an inclusive experience.

📊 Dataset Enhancements and Utility Features

New Features:

• Download All Payloads for Quick Replies and Buttons
  • This feature enables you to download all the payloads for quick replies and buttons in a text file. Use this file with the NLU Tester to ensure every button in your bot operates as intended.
  • To access this feature, go to Datasets > Intents and click on the "Download Buttons" button.

• Activate Exact Matching for Button Clicks
  • To streamline your Natural Language Understanding model, we’ve added a feature that allows you to bypass NLU for button click intents. When enabled, only 'Exact Matching' will be used, refining the performance of your NLU model.
  • To find this setting, open the intent in question and click on "Show more options."

Fixes:

• FAQ Search in Master Bot: Resolved the issue where FAQs from connected micro-bots were not always appearing in the master bot’s FAQ search.
• Search Result Indication: Fixed the issue where the answer search seemed to be ongoing, even when it had concluded with no results.
• Search Input Optimization: All search boxes have been updated to trim extra spaces, improving search accuracy.
• Error Message Refinement: Clarified an ambiguous error message that formerly read "Cast to ObjectId failed for value..." for a non-existent intent or entity.

🛠 Stability Improvements and User Experience Enhancements

Fixes in Builder

• Cross-Language Object Deletion: Fixed an issue where deleting an object in one language's graph could affect another language's graph.
• Initial Request Speed: Optimized the system to preload the trained model into RAM to eliminate first-request delays.

Fixes in Generative AI

• Training Status Feedback: Resolved occasional inconsistencies in displaying real-time training status.

📚 Updated Documentation

We've made several updates to our documentation to bring you the latest information and guidance. Here are some sections you may find particularly useful:

• Configuring Your NLU: This section now includes references to new classifiers and vectorizers, along with a comprehensive guide to help you optimize your NLU configurations.

• Understanding the Pros and Cons of Generative AI-Based Chatbots: This section provides an introductory overview of Smartly.AI's Generative AI capabilities.

• Implementing Generative AI in Your Chatbot 🤖: For a comprehensive guide and in-depth exploration, follow our step-by-step tutorial to get started with your first Generative AI bot.

• API Reference: Updated to include new custom code and bot messages endpoints.

🎉 In Conclusion

Thank you for being a part of the Smartly.AI community! We're committed to continually enhancing your experience and providing you with the tools you need to create exceptional conversational experiences. Your feedback is invaluable to us, so please don't hesitate to reach out with any questions or suggestions.

Here's to smarter bots and even better user experiences! 🌟