Setup Amazon Bedrock

amazon bedrock logo

Overview

🤖 Amazon Bedrock is a fully managed service that makes leading foundation models (FMs) from Amazon and third-party providers accessible via an API. It enables developers to:

✨ Key Features:

  • Access a choice of high-performing foundation models from leading AI companies through a single API
  • Build and scale generative AI applications securely with private customizations
  • Leverage enterprise-grade security and privacy controls
  • Create agents that can execute complex business tasks by combining foundation models with business logic
  • Fine-tune models with your own data while maintaining data privacy
  • Deploy models with high availability and performance at scale

🎯 Key capabilities include:

  • Text generation
  • Code generation
  • Image generation
  • Embeddings
  • Question answering
  • Summarization
  • Classification
  • Knowledge bases for RAG applications

🚀 Foundation models available through Bedrock include:

  • Amazon Titan
  • Anthropic Claude
  • AI21 Labs Jurassic
  • Stability AI
  • Cohere Command

🔧 Enterprise features:

  • Private customization and fine-tuning
  • Model evaluation and testing
  • Monitoring and observability
  • Security and access controls
  • Cost optimization

Retrieval Augmented Generation (RAG)

rag

Retrieval-Augmented Generation (RAG) is the process of optimizing the output of a large language model, so it references an authoritative knowledge base outside of its training data sources before generating a response. In this workshop we will use this technical to create an customer service agent based on our faqs document.

Benefits of RAG:

  • Improved Accuracy: RAG helps reduce hallucinations by grounding responses in verified knowledge sources

  • Up-to-date Information: Allows models to access current information beyond their training cutoff date

  • Domain Expertise: Can incorporate specialized knowledge bases and documentation for domain-specific responses

  • Cost Efficiency: More efficient than fine-tuning models since only the knowledge base needs to be updated

  • Data Privacy: Keeps sensitive data in controlled knowledge bases rather than embedding it in model weights

  • Verifiable Sources: Responses can be traced back to source documents for validation

  • Customization: Easy to customize model outputs by modifying the knowledge base

  • Reduced Training: No need to retrain or fine-tune models when information changes

  • Scalability: Knowledge bases can be updated independently of the model

  • Compliance: Better control over information sources for regulatory compliance

Prompt Engineering

Prompt engineering is the art of designing and refining prompts to elicit the desired response from an AI model. A prompt template will be:

  1. Clear Instructions
  • Be specific and explicit about the desired output
  • Break complex tasks into smaller steps
  • Include examples when helpful
  1. Context Setting
  • Provide relevant background information
  • Define the role/persona the model should adopt
  • Specify the format of the expected response
  1. Constraints and Parameters
  • Set boundaries for the response
  • Specify any limitations or requirements
  • Include validation criteria
Role: {specify the role/persona}
Context: {provide relevant background}
Task: {clear instruction of what to do}
Format: {specify output format}
Constraints: {list any limitations}
Examples: {provide sample input/output}

Example:

Role: Act as a technical documentation writer specializing in cloud computing
Context: Writing AWS service documentation for a beginner audience
Task: Create a step-by-step guide explaining how to launch an EC2 instance
Format: Numbered steps with bullet points for sub-steps, include relevant AWS console screenshots
Constraints: 
- Keep language simple and beginner-friendly
- Maximum 10 main steps
- Include security best practices
- Focus only on Linux instances
Examples:
Input: Need instructions for launching EC2
Output:
1. Sign in to AWS Console
   • Navigate to console.aws.amazon.com
   • Enter your credentials
2. Open EC2 Dashboard
   • Click "Services" dropdown
   • Select "EC2" under Compute
[etc...]

For more complex, you can add an history chat with an user to the prompt, but this required more token. In this workshop, for simple i just create an simple prompt to do something and then response via email.

Content

6.1. Request models
6.2. Create Knowledge base
6.3. Create Agent Alias