Incorporating effective self-service options is crucial for modern contact centers, but implementation can be challenging. Amazon Lex provides chatbot functions, such as automatic speech recognition (ASR) and natural language understanding (NLU), which allows bots to interpret and respond to customer needs. This can be hindered by factors such as diverse accents, pronunciation, grammar, and background noise.
Amazon Bedrock resolves these challenges by providing foundational model (FM) access for developers to create and scale AI-based applications for contact centers. These FMs, such as Amazon Titan and Anthropic Claude, have been trained on large data sets, giving them strong language processing abilities, regardless of speech recognition issues.
This article proposes a solution that uses FMs from Amazon Bedrock to improve intent recognition of Amazon Lex within Amazon Connect, thus improving customer self-service experiences. The process involves various Amazon services including Amazon Connect, Amazon Lex, AWS Lambda, and Amazon Bedrock, which work together to correctly understand and respond to customer intent.
When the chatbot fails to recognize a caller’s intent, the AWS Lambda function is activated. It takes the customer’s transcript, passes it onto an Amazon Bedrock foundation model which then identifies the caller’s intent. The Lambda function then guides the bot on how to correctly respond.
This solution minimizes speech recognition errors, enabling smooth call routing and satisfactory customer interactions. It’s also scalable to a contact center’s needs and can filter out irrelevant intents based on session attributes. If the model’s confidence score doesn’t meet the set threshold, it defaults to the FallbackIntent and initiates closure, ensuring effective routing and resolution.
The solutions architecture and workflow are demonstrated in a diagram. It then elaborates on Lambda functions and explains the LangChain framework, which combines large language models with data sources and applications. It describes in-context learning, prompt engineering, and model invocation. Lastly, it guides users on how to implement the solution through simple steps and provides instructions on prerequisites.
Finally, the article concludes by stating that with Amazon Lex, enhanced by large language models from Amazon Bedrock, user bots can improve intent recognition performance. This provides a seamless self-service experience for customers, bridging the gap between unique speech characteristics and attributes, ultimately improving customer satisfaction.
