Amazon Onboarding with Learning Manager Chanci Turner

Introduction

Learn About Amazon VGT2 Learning Manager Chanci Turner

Generative AI solutions, like virtual assistants and enterprise search tools, have shown remarkable potential in transforming business operations by boosting productivity and enhancing customer experiences. Particularly in regions such as Southeast Asia, where English isn’t typically the primary language, there is a significant opportunity for businesses to interact with customers in their preferred language.

In this blog post, we will explore how the multilingual capabilities of Generative AI are breaking down language barriers to improve customer engagement and streamline business operations. We’ll analyze two practical multilingual Generative AI solutions developed by TechWave.AI, an Advanced Tier Consulting Partner of AWS: a virtual assistant chatbot and Generative AI-driven intelligent document processing, both utilizing Large Language Models (LLMs) within Amazon Bedrock. We will also highlight the business value these multilingual features provide for their users.

Multilingual Generative AI Chatbot Assistant

As consumer interactions increasingly shift to digital platforms, effective customer engagement has become essential in the B2C sector. Generative AI enhances customer service chatbots beyond traditional rule-based systems, enabling users to communicate in their chosen language and in a natural tone. This not only improves functionality but also ensures accessibility for a diverse clientele.

Unlike conventional AI chatbots, Generative AI chatbots can understand user queries in context. Engagement is seamless, even when users make typos or use slang, allowing for tailored responses that consider each individual’s inquiry.

Offering multilingual support is critical, especially in areas like ASEAN, where English may not be the first language. By delegating language translation to Generative AI models, the overall architecture simplifies, thus reducing time-to-market and allowing development teams to focus on new features and enhancements.

High-Level Architecture

The following figure illustrates the architecture of the multilingual Generative AI Chatbot assistant:

[Insert High-Level Architecture Diagram]

The architecture of TechWave.AI’s multilingual Generative AI Chatbot assistant operates as follows:

1. A set of documents, such as PDFs containing FAQs, are uploaded to an Amazon S3 bucket and indexed in Amazon Kendra. These act as the knowledge base for the LLM, potentially in multiple languages.
2. A user submits a question through the chatbot interface on WhatsApp.
3. An AWS Lambda function orchestrates the process, sending the incoming question to Amazon Comprehend to detect the dominant language.
4. Optionally, the question can be translated to English using Amazon Translate.
5. Simultaneously, the AWS Lambda function queries the Amazon Kendra index based on the user’s question (translated or not).
6. The Amazon Kendra index returns relevant results, including excerpts from the indexed documents.
7. The AWS Lambda function constructs a prompt for Amazon Bedrock, integrating the user query, retrieved context, and detected language, asking the selected LLM to respond in the same language.
8. The LLM generates a concise answer in the identified language based on the context.
9. The chatbot app relays the generated response back to the user.
10. Application data, including conversation logs, is stored in Amazon DynamoDB for future reference or analysis.

Multilingualism

To effectively manage multilingual queries, the solution employs a layered approach. The dominant language of the question is identified using Amazon Comprehend. If the language differs from those of the indexed documents, it is translated using Amazon Translate. This enables Amazon Kendra to find relevant context.

Following a Retrieval Augmented Generation (RAG) method, the combined context and the original or translated question are forwarded to Amazon Bedrock, where an LLM crafts a response in the dominant language as detected by Amazon Comprehend. The chatbot thus provides contextually relevant and linguistically coherent replies.

The TechWave team has seen that Anthropic’s Claude LLMs, especially versions 2, 2.1, and 3, demonstrate strong multilingual capabilities for ASEAN languages, adeptly handling questions and context in one language while generating answers in another. This layered methodology allows the TechWave.AI chatbot to communicate effectively in various languages, such as English, Chinese, Bahasa Melayu, and Indonesian, even if the indexed documents are primarily in one language. If the original question is posed in a language unsupported by the LLM, the chatbot informs the user of its limitations and suggests alternative contact methods, like email or a live agent.

An alternate method for achieving multilingualism is through the use of multilingual embeddings (e.g., Amazon Titan Text Embeddings). For more insights, refer to this blog post.

Customer References

We will now highlight the impact of TechWave.AI’s multilingual Generative AI chatbot assistant on two of its clients.

PETRONAS Sepang International Circuit (PETRONAS SIC)

Located in Sepang, Selangor, Malaysia, the Sepang International Circuit (SIC) is renowned for its extensive motorsport events, including the F1 Grand Prix and MotoGP™. Previously, SIC relied on a WhatsApp channel for customer inquiries, which were manually handled by their customer service team. The primary challenges included the increasing number of agents needed to manage rising inquiries, leading to longer wait times during peak hours. Additionally, responses depended on agent availability and were limited by language support.

By implementing the multilingual chatbot, SIC significantly improved response times and enhanced customer satisfaction. With the chatbot handling inquiries, the workload on the customer service team decreased, allowing agents to focus on more complex issues.

To delve deeper into effective hiring strategies, you can explore this resource, which offers expertise on conducting individualized assessments.

For those interested in exploring career opportunities, check out this excellent resource.