What is Amazon Bedrock and How Do You Use It?
What is Amazon Bedrock and How Do You Use It?
Generative Artificial Intelligence (AI) is growing with almost unbelievable speed and almost every business is considering or even using generative AI. AWS (Amazon Web Services), as usual, isn't getting left behind, and AWS events always run great sessions about AI services.
Of the long list of AI services that you can use, we have selected Amazon Bedrock today because it's one of the most anticipated.
In this blog post, we'll take a look at what Amazon Bedrock is, its features, its use cases and some simple examples in real-world scenarios, and how to use it in the AWS console.
What is Amazon Bedrock?
Amazon Bedrock is one of the fully managed AWS services that offers a great set of features that leverage major FMs (Foundation Models) from leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon.
Currently Amazon Bedrock is available in multiple regions, including US-EAST 1, EU-CENTRAL 1 and other Asian, US and EU regions. You should first check the region that you need to use, because some regions don't have access to all FMs.
Amazon Bedrock is currently the quickest way to build and scale generative AI applications in the AWS environment.
Bedrock is serverless, so you can get started quickly without the need to manage any servers, as well as privately customize it with your data and easily integrate and deploy it into your applications. And because it's a regional service, you can deploy it in the region closest to you or your users.
It also has some great advantages for those who already use AWS services, because you get a secure and simple connection between your current solutions to Bedrock.
Benefits of using Amazon Bedrock
- Ease of access - Bedrock simplifies access to powerful AI models, meaning businesses can quickly deploy sophisticated AI without needing deep expertise in its development.
- Scalability - By leveraging AWS infrastructure, AI applications built with Bedrock can scale as needed, handling large volumes of requests efficiently.
- Cost-effectiveness - Since Bedrock offers pre-trained models, companies save on the costs and resources associated with training large models from scratch.
Amazon Bedrock's key features and capabilities
The capabilities of Amazon Bedrock include the following:
- Foundation models that include a choice of base FMs and customized FMs.
- Playgrounds for chat, text, and images with quick access to FMs for experimentation and use through the console, now with multimodal support.
- Builder tools like knowledge bases, agents, prompt management and prompt flows with governance and memory capabilities.
- Safeguards such as watermark detection and guardrails with enhanced injection detection.
- Model evaluation for assessment and deployment of the right FMs.
- Model customization with reinforcement fine-tuning for automated accuracy improvements.
Let's take a deeper look at the features that you will probably leverage when you want to start using Amazon Bedrock.
Foundation models
Pre-trained models: These models are already trained on vast amounts of data, making them powerful tools for a wide range of AI tasks without needing extensive additional training.
Amazon Nova models: Amazon's own generation of foundation models designed for enterprise applications. Nova 2 Omni (in preview) is the first multimodal model on Bedrock that can understand text, images, video, and speech while generating both text and images. Nova 2 Sonic delivers natural multilingual voice interactions with speech-to-speech capabilities.
Custom frontier models with Nova Forge: Organizations can now build custom frontier models using Nova Forge, allowing you to embed domain expertise by blending proprietary data with Amazon Nova-curated datasets without traditional barriers of cost and compute.
Customization and fine-tuning: Users can fine-tune these models on their own data to better suit their specific needs. New reinforcement fine-tuning delivers 66% average accuracy gains over base models using feedback-driven training without requiring large labeled datasets or deep ML expertise.
Reinforcement Fine-Tuning (RFT)
Amazon Bedrock now supports reinforcement fine-tuning, helping you improve model accuracy without needing deep machine learning expertise or large sums of labeled data. Amazon Bedrock automates the reinforcement fine-tuning workflow, making this advanced model customization technique accessible to everyday developers. Models learn to align with your specific requirements using a small set of prompts rather than the large sums of data needed for traditional fine-tuning methods, enabling teams to get started quickly. You can define reward functions using rule-based graders or AI-based judges, optimize for both objective tasks (code generation, math reasoning) and subjective tasks (chatbot interactions, instruction following), and train directly through the console or APIs without your proprietary data ever leaving AWS's secure environment.
Model evaluation
One of the major steps in selecting the correct AI model is its evaluation. This process can be difficult, but Amazon Bedrock already has a simple-to-use evaluation tool included in the AWS Management Console.
Playgrounds
This feature closely resembles classical generative AI models like ChatGPT or Gemini. It allows you to use chat, image, and text playgrounds to test FM models or use them as generative AI in the simplest way possible.
Chat playground
This helps you test the models for chat to assess how they interact with your prompts or to just use them as normal generative AI models.
Text playground
If you need to generate text based on prompts, you can also test the specific models for this task.
Image playground
And yes you can also test out a couple of models for image generation or even use the images to your liking if you need to.
Multimodal playground with Nova 2 Omni
With Nova 2 Omni now in preview, you can test models that simultaneously process text, images, video, and audio inputs. This unified multimodal approach enables comprehensive understanding across all content types in a single request, opening new possibilities for complex AI applications.
Builder tools
Knowledge bases
This is the quickest way to create a "database" for the FMs to retrieve information from. You can, for example, use a set of documents from the S3 buckets, which are accessed by the FMs when you prompt them. Essentially, knowledge bases make the FMs smarter by giving them access to detailed, business-specific information that you have already. Knowledge bases now integrate with Amazon S3 Vectors for cost-optimized vector storage, reducing costs by up to 90% compared to traditional vector databases.
Agents
Amazon Bedrock agents can interact with the FMs, other AWS services, and external systems to perform tasks, make decisions, or handle requests. Agents now feature AgentCore governance with policy controls, quality evaluations, and episodic memory for learning from past interactions.
HERE IS A SIMPLE EXAMPLE OF HOW TO USE AMAZON BEDROCK AGENTS:
Imagine you run an online store and want to automate customer support for order tracking.
1. Set up the agent: You create a Bedrock agent that can interact with an AI model trained to understand customer inquiries about orders.
2. Connect to data sources: The agent is connected to your order database and shipping service.
3. Process customer requests: When a customer asks, "Where's my order?" the Bedrock agent uses the selected model to understand the request, and fetches and retrieves the latest tracking information from the shipping service.
4. Respond to the customer: The agent then generates and sends a response such as "Your order was shipped and is expected to arrive in two days."
Model Evaluation
With model evaluation on Amazon Bedrock, you can easily assess, compare, and choose the best foundational model for your needs. Evaluation capabilities help ensure you select the right model for your specific use case before deployment.
Safeguards
Watermark Detection
Watermark Detection for Titan Image Generator identifies invisible watermarks embedded in every image the model creates. This watermark helps prevent the spread of disinformation, supports copyright protection, and tracks content usage.
Guardrails
Guardrails for Amazon Bedrock assesses user inputs and model responses according to specific use-case policies, adding an extra layer of protection regardless of the foundational model being used. You can, for example:
- Detect and block harmful user inputs and model responses like hate and violence.
- Create a specific topic that your FMs will not answer.
- Add specific blocked words.
Amazon Bedrock use cases and examples
The best way to understand how Amazon Bedrock could be used in your business is by going through a set of use cases and examples from a real-world perspective.
Text generation
Craft original content like short stories, essays, social media posts, or website copy.
Real-world use case: A marketing team uses text generation to quickly create engaging social media content for product launches.
Virtual assistants
Develop assistants that can understand user requests, break down tasks, engage in conversations to gather information, and take actions to complete those tasks.
Real-world use case: A customer service chatbot is designed to handle inquiries, such as processing returns or tracking orders, reducing the need for human agents and improving response times.
Text and image search
Find and compile relevant information to answer questions and provide recommendations from extensive text and image data.
Real-world use case: A legal firm uses text and image search to quickly locate relevant case law and visual evidence from a vast database, helping lawyers prepare for court more efficiently.
Video understanding and search
Analyze and search video content for insights, patterns, and relevant information.
Real-world use case: A media company uses video foundation models like TwelveLabs Marengo 3.0 to understand video content across dialogue, gestures, movement, and emotion, enabling semantic search across millions of hours of video without frame-by-frame analysis.
Text summarization
Generate concise summaries of lengthy documents, such as articles, reports, research papers, technical manuals, and books, to quickly capture key information.
Real-world use case: A researcher uses text summarization to condense multiple lengthy research papers into brief overviews, allowing them to quickly review key findings without reading every detail.
Image generation
Create realistic, visually appealing images for ads, websites, presentations, and more in no time.
Real-world use case: A graphic designer generates high-quality images for an advertising campaign, speeding up the design process and enabling rapid iterations based on client feedback.
Custom domain models
Build custom frontier models that embed domain-specific expertise without massive cost or compute requirements.
Real-world use case: A financial services firm uses Nova Forge to build a custom model combining public market data with proprietary trading insights, replacing multiple specialized models with a single, more accurate solution.
Guardrails
Set up safeguards tailored to your application's needs and AI policies for responsible usage.
Real-world use case: A healthcare application uses guardrails to ensure that generated responses to medical queries are accurate, safe, and comply with industry regulations, protecting both users and the organization.
How to use Amazon Bedrock?
Because of the variety of use cases, it's hard to select the best way to start using Amazon Bedrock. Though it's available through AWS Management Console and API, it's really easy to start in the Console. A probable first step for everybody is to enable the model access in the Bedrock Console, because by default you don't have access to any FMs. It's really simple: you just enable the models you need. There shouldn't be any associated costs with doing so.
You can then access these models with all the features and functions previously described. I recommend first experimenting with the playgrounds to understand the models and how to interact with them. Once you're ready to move beyond experimentation, you can explore reinforcement fine-tuning through the console to customize models for your specific use case, or try Nova Forge to build custom domain models. After that, you can follow some of the excellent workshops that AWS already offers for Amazon Bedrock.
If you want to play with Amazon AI, then also visit the AWS PartyRock page. It's a really fun way to start learning!
Amazon Bedrock security and responsible AI
So let's answer some common questions regarding the security of Amazon Bedrock, such as where is my data stored? Is my data stored and transferred securely and is it encrypted? Can I connect to Amazon Bedrock securely from an on-premises environment? Can I restrict access in the AWS environment to this service?
- Your provided data is not used to improve or enhance the base FMs. Your data is not shared with any model providers. With AgentCore governance, you can additionally control which data your agents access and how they use it.
- As a regional service, Amazon Bedrock offers possibilities for storing data where you need it. The data that you store, including prompts, attachments that you use for prompts, AI responses and customized models remain in the selected region.
- All data is also encrypted in transit via TLS 1.2 and at rest with service-managed AWS Key Management Service (AWS KMS) keys.
- If you need to connect on-premises environments to Amazon Bedrock you simply leverage AWS PrivateLink.
- With AWS Identity and Access Management (IAM), you can manage who has access to your customized foundational models. You can decide which specific models and services are accessible and control who can log in to the Amazon Bedrock console. AgentCore Policy adds an additional layer where teams define agent tool access using natural language.
Conclusion
Amazon Bedrock continues to evolve as the quickest way to build and scale generative AI applications in the AWS environment. With access to fully managed models from leading providers, advanced customization through reinforcement fine-tuning, new multimodal capabilities with Nova 2 Omni, and enterprise governance through AgentCore, organizations can confidently deploy AI applications from prototypes to production.
So this was our first blogpost about Amazon Bedrock. We have described what it is, highlighted some of its features, explored simple use cases for multiple industries and explained how to start using it right now in the AWS Console.
An AWS Solutions Architect with over 5 years of experience in designing, assessing, and optimizing AWS cloud architectures. At Stormit, he supports customers across the full cloud lifecycle — from pre-sales consulting and solution design to AWS funding programs such as AWS Activate, Proof of Concept (PoC), and the Migration Acceleration Program (MAP).
