Amazon Bedrock - Amazon Bedrock Tutorial

- [Instructor] Before we get started building generative AI applications with Amazon Bedrock, we're going to quickly review a few fundamental technologies and concepts that will be used in this course. Amazon Bedrock is a fully managed service, which offers a choice of high performing foundation models and capabilities, allowing you to build generative AI applications easily while maintaining privacy and security. One of the key features of Amazon Bedrock is the ability to choose from a variety of underlying foundation models, depending on your use case. These can all be accessed via a single API. Examples of models like Titan from Amazon and Claude from Anthropic help with question and answering, text summarization, and information extraction. Other models like SDXL from stability AI, and even Amazon Titan can help with text-to-image generation. New models are constantly being added to Amazon Bedrock, so you'll be sure to find one that supports your generative AI use case. Although you can use Base FMs out of the box, there is the option to customize them with your own data. Amazon Bedrock was built prioritizing security, so any data you use is private and will not be shared back with AWS or model providers. Any customer data that is used as part of training, fine tuning, or responses, remains in the region where it was created and is not made available to the underlying AWS engine. Amazon Bedrock also has the ability to configure agents, which extend capabilities to help perform orchestration and retrieve information in real time from external data sources. Agents allow you to dynamically invoke APIs to help you complete tasks as part of a sequence of events in a chain for your application. Vector stores are an important part of generative AI, they're used to perform similarity searches that can help provide additional context to the prompts to make more accurate responses. Amazon Bedrock has built-in support for vector store integration with knowledge bases. You are able to embed and index documents for retrieval using knowledge bases, which uses Amazon OpenSearch under the hood. When accessing your models, you can use Amazon Bedrock on demand, where you pay as you go with no commitment. The other option is provision throughput, where you reserve capacity at a fixed cost, which is a better option for production workloads. This guarantees higher limits with flexible commitment terms. That's a lot of features that Amazon Bedrock provides. But next, let's discuss one of the most popular frameworks for building generative AI applications, which is LangChain.