Amazon Bedrock + DeepSeek

Set up and run DeepSeek-R1 in Amazon Bedrock

Goals

Install DeepSeek-R1 in Amazon Bedrock and try DeepSeek. And, Understand Amazon Bedrock's capabilities to jumpstart AI adoption.

About Amazon Bedrock

Amazon Bedrock is a fully managed service that offers high-performing foundation models (FMs) from leading AI companies and provides a unified API and a Bedrock Studio.

• 170+ Foundation models (FMs) from leading companies and the marketplace. See below few list of Amazon Bedrock foundation models
• Import 3rd party models into Bedrock
• Leverage readymade Foundation models (FMs)
• Build custom Foundation Models
• Extend existing Foundation Models with domain-specific Knowledge bases

Install DeepSeek-R1 in Amazon Bedrock

1. Open the AWS console, go to Amazon Bedrock, and provision the DeepSeek-R1 Model
2. Bedrock Knowledge Bases
3. Provision Amazon Bedrock Guardrails
4. Amazon Bedrock Studio/Amazon Bedrock IDE
5. Bedrock Prompt Management
6. Bedrock Agents
7. Bedrock Playgrounds: Chat / Text, Image / Video
8. Bedrock Flows or Workflows
9. List of Amazon Bedrock models

Step-1: Open the AWS console, go to Amazon Bedrock and provision DeepSeek-R1 Model

• Open the Amazon AWS console or use any AWS SDK tools you are currently using for provisioning. I will be using AWS console for a quick turnaround on this post.
• Search "Bedrock" at the top. and, select Amazon Bedrock
• Select "Marketplace deployments" at the left navigation bar
• Click on the "View Model Catalog" button
• Search "DeepSeek" which shows "DeepSeek-R1" as a search result. Click on the "DeepSeek-R1" model to deploy the model.

Step-2: Amazon Bedrock Knowledge Bases

Amazon Bedrock Knowledge Bases is a fully managed service that enhances generative AI applications by integrating internal domain-specific databases via Retrieval-Augmented Generation (RAG) workflows. It allows AI models to access and retrieve relevant domain-specific proprietary data, improving the accuracy and relevance of responses.

• Supports structured and unstructured data; Text (documents, knowledge bases, manuals, FAQs), Multimodal Data (charts, images, tables, diagrams), and AWS Data Services (Amazon S3, Amazon OpenSearch, and Amazon RDS/Databases, etc)
• Retrieval-Augmented Generation (RAG) Support
  • Enhances AI responses by retrieving information from connected data sources (e.g., databases, documents, and knowledge repositories).
  • Reduces hallucinations (incorrect AI-generated content) by grounding responses in real data.
• Vector Embeddings & Indexing
  • Converts textual and visual data into vector embeddings (numerical representations of data).
  • Enables semantic search, allowing AI models to find relevant data efficiently.
• Custom Prompts & Query Optimization
  • Allows users to customize AI-generated responses by configuring retrieval parameters and prompts.
  • Features advanced parsing, chunking, and query reformulation for better accuracy.
• Fully Managed RAG Solution
  • No infrastructure management is required—AWS handles indexing, retrieval, and scaling.
  • Easy integration with Amazon Bedrock and foundation models like Claude, Llama, and Titan.

Step-3: Provision Amazon Bedrock Guardrails

Amazon Bedrock Guardrails allows you to implement AI safety. It has various content filtering features to add safety and reduce hallucinations.

• Content filters
• Denied topics
• Word filters 
• Sensitive information filters
• Hallucinations filter via contextual grounding check
• Image content filter

You can apply Privacy and Security with Guardrails

Step-4: Amazon Bedrock Studio/Amazon Bedrock IDE

Amazon Bedrock Studio lets users in your organization easily experiment with Amazon Bedrock models and build applications. You can

• Create workspace and projects
• Add and Manage collaborative users
• Manage Prompt engineering and workflows

Step-5: Amazon Bedrock Prompt Management

Amazon Bedrock Prompt Management allows you to optimize the prompts (input) for specific use cases and models. It helps you to optimize textual input to a Large Language Model (LLM) to obtain desired responses. It helps you to

• Design prompts and create prompt templates
• Create and Manage (View, Modify, Test, Optimize, Delete) a set of prompts
• Routing prompts across different foundational models within the same model family

Step-6: Amazon Bedrock Agents

Amazon Bedrock Agents allows you to create front-end software and orchestrate interactions between backend foundation models (FMs), data sources, software applications, API calls, invoke knowledge bases, and user conversations. It helps you to quickly create a front-end agent to interact with backend stacks. You can

• Build and modify agents
• Create actions and manage in ActionGroup
• Create and manage ActionGroups
• Enable or disable Multi-agent collaborations
• Enable or disable user input
• Augment response generation for your agent with the knowledge base
• Retain conversational context across multiple sessions using memory
• Generate, run, and test code for your application by enabling code interpretation
• Deploy, test, and troubleshoot agent behavior

Step-7: Amazon Bedrock Playgrounds: Chat / Text, Image / Video

Amazon Bedrock provides readymade Playgrounds UI to connect and test Foundation Models (FMs). You can test and interact with Amazon Bedrock foundation models by using the following four playgrounds:

• A text playground.
• A chat playground.
• A voice chat playground.
• An image playground.

Step-8: Amazon Bedrock Flows or Workflows

Amazon Bedrock Flows (workflows) allow you to build end-to-end solutions by linking prompts, foundational models, and other AWS services quickly. You can create, modify, or delete workflows quickly.

Step-9: List of Amazon Bedrock models

Here are some leading Foundation models offered by Amazon Bedrock

Note: I focussed mostly on Amazon Bedrock in this post since you can deep dive into DeepSeek's Install DeepSeek AI Model & Start Your Own Chat

Conclusion

With Amazon Bedrock tools, you can leverage existing Foundation Models (FMs), and enhance or customize FMs. It provides all the tools to build your next Foundation Model (FM) or related solutions.

References

Install DeepSeek AI Model & Start Your Own Chat

Set up and run DeepSeek-R1 on AWS cloud with EC2, and Try DeepSeek Queries. You can do the same in your own servers, IBM cloud, Oracle Cloud, Google cloud, Azure cloud or personal laptop. But, I will choose AWS in this post for simplicity. Goals Install DeepSeek-R1 in AWS cloud,

expectations.aiRavi Kalidindi

Amazon EC2 Capacity Blocks for ML Pricing – AWS

Find pricing information and pricing examples for Amazon EC2 Capacity Blocks for ML

Amazon Web Services, Inc.