AWS has released a sample project showing how businesses can deploy an Amazon Lex bot that leverages a pre-trained, open-source LLM.
Several other conversational AI vendors have taken similar steps, integrating with LLMs like ChatGPT.
In doing so, these providers have enabled customers to build bots using natural language alone, simulate conversations for testing, and adapt the tone of replies based on customer responses.
Yet, AWS’s demo – as showcased in a recent blog – highlights how Lex users can implement two very different use cases.
The first use case revolves around what AWS calls a "custom memory manager."
What this does is allow the bot to recall previous customer conversations to maintain the interaction’s context and speed.
The second use case enables Lex to better handle bot fallbacks.
A fallback occurs when the user’s input does not match any of the intents that the company has trained the bot to handle.
The LLM improves that fallback by analyzing what the customer said/wrote and looking to see which intent it best matches.
As an example of this, consider a florist. If a customer asks: "I’d like to order chocolate," the LLM can identify that the customer wants to order something.
It could then ask: "Did you mean: I’d like to order flowers." That is most likely what the customer meant in that scenario – which helps pull the conversation in the right direction.
In bringing these use cases to life, AWS leveraged its Lambda solution and an open-source LLM from Amazon SageMaker JumpStart.
Jumpstart is its portfolio of machine learning models that businesses can test and tune.
