Amazon Bedrock Guide : Generative AI with AI Agents

Unlock the Power of Amazon Bedrock to Build AI-Powered Applications

Welcome to Mastering Amazon Bedrock, a comprehensive course designed to help you harness the power of AWS Bedrock’s tools and services. Whether you're a beginner or an experienced developer, this course will take you step-by-step through concepts, configurations, and hands-on exercises that showcase the potential of AWS Bedrock in building intelligent applications.

What You’ll Learn:

• Knowledge Bases (KB): Dive deep into the concept of vector embeddings and retrieval-augmented generation (RAG), essential for optimizing large-scale AI applications. Learn how to configure Knowledge Bases and integrate them seamlessly with other AWS Bedrock tools using practical examples to solidify your understanding.

  • RAG with Amazon Bedrock - We will use Anthropic Claude Model with OpenSearch Serverless as vector storage to perform the RAG operations

  • RAG with Open Source - We will also use OpenAI's ChatGPT model with in memory vector storage to perform RAG operations

  • Retrievers - RAG pattern relies heavily on retrieval. There are many ways to retrieve data for summarization. We will learn and explore about different ways to retrieve the contents. Followed by a hands-on activity   

• AI Agents: Master the configuration of AWS Bedrock agents to streamline AI workflows. Gain hands-on experience in implementing action groups, handling parameters, and orchestrating requests effectively to Knowledge Bases. Understand how agents serve as the backbone of dynamic and intelligent AI interactions. We will cover 2 use cases of AI Agents.

  • Multimodal Nutritional AI Agent - We will use Open Source components like Haystack, FastRag, HuggingFace with Multimodel modal Phi-3.5-vision-instruct to run multi Agentic use case. We will also cover multi agentic Tools with Multi-Hop and ReAct Prompt.

  • Multi-Agentic Travel AI Agent - We will use Open Source framework - CrewAI and OpenAI ChatGPT model with planning and reasoning ability using Tools with Multi-Hop and ReAct Prompt.

• Prompt Management: Develop expertise in creating, managing, and optimizing prompts to fine-tune AI responses. Explore the use of variables and strategies for effective prompt engineering, a critical skill for delivering customized user experiences in AI applications.

• Flows: Learn to build advanced workflows by integrating Knowledge Bases, AI Agents, and Prompts. Flows allow you to design seamless interactions and manage complex application logic, ensuring efficient and scalable AI solutions.

• Hands-On Lab: Apply your knowledge through hands-on labs that walk you through building end-to-end solutions. Combine Knowledge Bases, AI Agents, Flows and Prompts to create practical, real-world AI applications that solve complex problems.

• Guardrails: Understand the importance of security and compliance in AI systems. Learn how to implement robust guardrails to ensure your applications adhere to best practices, remain reliable, and mitigate risks effectively. We will cover different Guardrails Topics like Hallucination, Prompt Injections and take a deep dive into each one of them.

  • Guardrails with Amazon Bedrock - We'll do a hands-on Guardrails(text, image) on Bedrock platform.

  • Guardrails with Open Source tools - We will also do a hands-on Guardrails with Open Source models like Prompt Guard (Llama Family), Phi3 Hallucination Judge from HuggingFace to detect Prompt Injection and Hallucination respectively on a Google Colab notebook.

• Evaluators:  Evaluate, compare, and select the foundation model for your use case with Model Evaluation. Prepare your RAG applications for production that are built on Amazon Bedrock Knowledge Bases or your own custom RAG systems by evaluating the retrieve or retrieve and generate functions.

  • We will cover topics like LLM-As-A-Judge, Context Relevancy using Amazon bedrock platform and open source tools

• Batch Inference: With batch inference, you can submit multiple prompts and generate responses asynchronously. Batch inference helps you process a large number of requests efficiently by sending a single request and generating the responses in an Amazon S3 bucket.

• Model Fine Tune: We will fine-tune a pre-trained foundation model to take advantage of their broad capabilities while customizing a model on your own small, corpus.