What’s Changing?
With the Natural Language system, when the user asked a question or made a request, our NLU model would identify what they meant by matching the input to pre-defined “intents.” For example, if they said, “What’s the weather today?” the system would recognize it as a weather-related intent and respond accordingly.
With this new RAG system, we still rely on predefined intents, but the way we handle and process your utterances has become significantly more advanced. RAG uses embeddings to better understand the context and meaning behind the words, making it more flexible and accurate when matching them to the right intent. This means faster, more efficient intent recognition and an improved ability to handle diverse or complex utterances.
Why This Is Better for You
Faster Training And Interaction
RAG training and interactions are significantly faster and more efficient than traditional NLU systems. This means quicker training times for our agents and faster, more accurate classifications when matching your intents, ensuring a smoother conversational experience. For example, an agent with 37 intents and 305 utterances now trains about 20 times faster: in about 1 second.
Automatic agent training
The advanced training speed unlocked by the new RAG system means that explicit training is no longer needed. Test your agent and training will happen automatically.
Greater Understanding of Complex Questions
Embeddings are a way of representing language in a rich, context-aware format. The new RAG system leverages embeddings to understand the deeper meaning behind the words, even if they're phrased differently. Users can also ask detailed, complex questions, and the system will be able to understand the underlying context better. For example, a sentence like, “What’s it like outside, is it going to rain later?” might not have matched perfectly with an intent for “weather today” in the NLU system. Now, embeddings will understand the utterance, "is it going to rain later," as a query into the current weather and match to the correct intent.
A More Natural Experience
Ultimately, this upgrade means a smoother, more conversational experience for you and your customers. Whether you type casually, use slang, or ask detailed questions, our system is designed to understand accurately and efficiently.
What Does This Mean for You?
From your perspective, the change should feel effortless—no need to change how you build, test, and launch AI agents. What you will notice is:
- fewer misunderstandings;
- more accurate responses to your questions, and;
- a faster, more personalized experience overall.
However, since some user utterances have been specifically tailored for the NLU system, you may notice slight differences when using the new RAG system. In some cases, minimal adjustments to the phrasing of utterances may be needed to optimize performance with the new approach.
To make this transition smoother, we will keep both services running for a period of time. This will give you the necessary opportunity to explore the new system, test it thoroughly, and make any adjustments to your agents as needed. You can switch between the NLU and RAG systems in intent classification settings within the Intents CMS.
By adopting cutting-edge technology like RAG with embeddings, we’re making sure our system can grow and adapt alongside your needs, giving you the best experience possible. We’re thrilled to bring you this improved experience, and we’d love to hear your feedback as you interact with the new system. As always, your input helps us continue to innovate and serve you better.
.svg)