SUVA v23.09 Release Notes

SUVA's v23.09 release is available to all users from 28 September. This release can be summed up as a lean yet pragmatic one. With this release, SUVA chatbot becomes all the more capable to better manage user interactions. Along with some small-time changes, there's an important upgrade. Let's discuss the details and highlights of this release.

LLM-Powered Chatbot Interactions

When LLM integration is enabled in SUVA, then all of its interactions are managed via large language model. This is a big step up from earlier, when only the fallback responses were generated using the LLM integration. All the chatbot-managed interactions mean more human-like chatbot responses.

Related doc: LLM Integration Usage

Note. The randomness of the chatbot responses depends on Temperature Setting, which was announced in v23.07 release.

Profanity Layer to Detect Flagged Words

To detect and disregard bad or flagged words, there is a profanity layer has been introduced. Implementing the profanity layer in a chatbot involves adding a filtering mechanism to prevent the chatbot from generating or responding with inappropriate or offensive language. This is a crucial step to maintain a positive and respectful interactions with users.

Queries detected by the profanity layer would not get any response from the chatbot.