Short course (video) introducing how to build serverless LLM-based applications using Amazon Bedrock. Instructors (Andrew Ng and Mike Chambers) present a workflow that detects new customer service call recordings, uses ASR to transcribe audio, then triggers LLM-based summarization within an event-driven, serverless architecture to simplify deployment and scaling.
One might think that building and running an LLM application, say taking audio from a customer call center, transcribing it using automatic speech recognition, summarizing the call using an LLM, and so on requires setting up either a locally hosted or a cloud hosted compute server to run all this. But there's another potentially easier way to build such applications using a serverless architecture, and that's what you learn to do in this course. One fun bonus if you haven't used automatic speech recognition or ASR much before, you also learn how to do that all in the context of a serverless AI architecture. Specifically in this course you learn to build a system that will automatically detect when a new customer service call comes in, and then automatically run a complex multi step workflow to generate and store a summary and will be doing this on top of Amazon Bedrock the instructor, Mike Chambers is a developer advocate at AWS specializing in generative AI. Mike has helped a lot of AWS customers build a huge range of applications and so he's seen a lot of what actually does and doesn't work in practical applications. He's also an instructor of a very popular Coursera course on generative AI with large language models. Thanks Andrew. It's great to be here. In this short course you'll learn how to deploy your LLM based application into a production grade event driven architecture, all using serverless technologies. So why serverless? The usual deployment process would be fairly involved, including spinning up dedicated machines or even Docker containers, installing complex dependencies, and then there's the ongoing maintenance, adding security patches and so on. Management of this system would normally be another full time job. By going serverless you don't have any infrastructure to manage and you can deploy your LLM based app in days instead of weeks and also scale it up to millions of users without needing to worry about server capacity or scaling. The key to implementing an AI workflow using a serverless architecture is building an event driven architecture which uses events to trigger computational steps in the pipeline. For example, whenever a new customer service recording gets uploaded to a storage location, we call that an event, which then automatically triggers the process of using ASR to generate a text transcript of the audio. When the transcript is complete, that generates another event which leads to triggering the LLM to summarize the conversation and so on through the steps of the workflow. You'll use Amazon Bedrock, which lets you access a wide range of LLMs from a number of different providers and integrate them with other serverless cloud based services. So I hope you enjoy the course that's great. Would you like some tea? Oh, we want to be serverless in this course.
