Using multiple agents for scalable generative AI applications (AIM304)

Introduction

  1. This session is focused on multi-agent collaboration using Amazon Bedrock, a fully managed service that offers a broad set of high-performing models from leading industry providers.
  2. The presenters are Michael L, the product lead for Bedrock agents, and Mark Roy, the principal architect at Amazon Bedrock.
  3. They are joined by Hio Zerker, a principal architect at Northwestern Mutual, who will share a customer case study.

Key Takeaways

  1. Single Agent Limitations: As agents scale and take on more responsibilities, they can become less effective, decreasing task success, accuracy, and overall goal achievement. This is a common design pattern observed in the industry.
  2. Benefits of Multi-Agent Collaboration: Leveraging multiple, specialized agents orchestrated by a supervisor agent can improve problem-solving accuracy, scalability, and resilience. It also enables reuse and composability of agents built by different teams.
  3. Amazon Bedrock Multi-Agent Collaboration: The solution provides the following capabilities:
    • Easily assemble a set of smaller, focused agents and knowledge bases.
    • Utilize a supervisor agent to plan and execute workflows across the collaborating agents.
    • Provide unified conversations and intent-based routing across agents.
    • Offer observability and traceability across the multi-agent flows.
    • Deliver the same trusted security, privacy, and data governance as other AWS services.

Unified Customer Experience

  1. The presenters demonstrate how multi-agent collaboration can be used to create a unified customer experience, where a supervisor agent routes the user's requests to the appropriate specialized sub-agents, providing a seamless and personalized experience.
  2. The routing is handled by the supervisor agent through intent classification, without the need for complex orchestration code.

Automating Complex Processes

  1. Multi-agent collaboration enables the automation of complex processes by breaking them down into smaller, specialized tasks handled by individual agents.
  2. The supervisor agent can plan and execute a multi-step workflow, leveraging the capabilities of the various sub-agents, and return a consolidated result.
  3. This approach can be applied to various use cases, such as investment portfolio analysis, insurance claim processing, and developer productivity tools.

Customer Case Study: Northwestern Mutual

  1. Hio Zerker shares how Northwestern Mutual leveraged Amazon Bedrock Agents to create a chatbot-based solution for their customer support operations.
  2. The solution includes various specialized agents, such as for user management, documentation retrieval, and pipeline failure analysis, orchestrated by a supervisor agent.
  3. Key lessons learned include the importance of well-curated data and documentation, cross-region inference for stability and performance, and the need for observability and feedback mechanisms to ensure a good user experience.

Call to Action

  1. The presenters provide a QR code to access an open-source library of Amazon Bedrock Agent samples, including both single-agent and multi-agent collaboration examples.
  2. The audience is encouraged to try out these samples, experiment with creating their own agents and multi-agent systems, and provide feedback to the presenters.

Your Digital Journey deserves a great story.

Build one with us.

Cookies Icon

These cookies are used to collect information about how you interact with this website and allow us to remember you. We use this information to improve and customize your browsing experience, as well as for analytics.

If you decline, your information won’t be tracked when you visit this website. A single cookie will be used in your browser to remember your preference.

Talk to us