site-logo

JAWS PANKRATION 2024

AWS GenAI: Chatbot with Bedrock

Lv200

Lv200

8/24/2024 15:20 (UTC)

Session Info

We will explore AWS GenAI's Amazon Bedrock with an example to create a chatbot, and demonstrate how to use this service for integrating Gen AI models to enable intelligent and efficient interactions.

Abel  Lopez

Abel Lopez

- AWS Community Builders -



Session Category
etc


AWS Services

Session Materials


    Session Summary (by Amazon Bedrock)
      Abel López, a mining engineer and cloud professional, presents on building a chatbot using Amazon Bedrock, AWS's generative AI service. He highlights Bedrock's ability to simplify the use of foundation models through a single API, offering customization, retrieval-augmented generation, and security features. López discusses various foundation models available in Bedrock, including Claude, Titan, and others. He emphasizes that there's no single model for all purposes, and users can combine different models based on their needs. The presentation focuses on using Claude, specifically the Haiku version, which offers a balance of performance and cost. López explains different levels of model customization, including prompt engineering and retrieval-augmented generation. The chatbot creation process involves: 1. Configuring a knowledge base using S3 bucket data 2. Embedding the data into a vector database 3. Setting up an agent to access the database and generate augmented responses López demonstrates the process using the AWS console, creating a knowledge base with S3 data, selecting Titan for embeddings, and choosing OpenSearch as the vector database. He then creates an agent using Claude 3, configuring it with instructions and connecting it to the knowledge base. The resulting chatbot can answer questions based on the ingested information. López briefly mentions the option to add guardrails for enhanced security and compliance. The presentation concludes with a demonstration of the chatbot answering questions about a specific event (Dios Creation) using the information from the knowledge base.

    ©JAWS-UG (AWS User Group - Japan). All rights reserved.