🧠 Amazon Bedrock GenAI Workshop

No-Code RAG Pipeline & AI Agent with Code Interpreter

Build an end-to-end Retrieval-Augmented Generation (RAG) pipeline and a code-interpreting AI Agent using Amazon Bedrock — entirely through the AWS Console, no coding required.

📌 Workshop Source: AWS Bedrock Workshop

---

📑 Table of Contents

• Architecture
• Prerequisites
• What You'll Build
• Section 1 — RAG Knowledge Base Pipeline
  • 1.1 — Create Vector Store (OpenSearch Serverless)
  • 1.2 — Upload Documents to S3
  • 1.3 — Create Bedrock Knowledge Base
  • 1.4 — Test the Knowledge Base
  • 1.5 — Build RAG Flow (Visual Builder)
• Section 2 — AI Agent with Code Interpreter
• Resources

---

🏗️ Architecture

  ┌───────────────────────────────────────────────────────────┐
  │                       Amazon Bedrock                      │
  │                                                           │
  │  ┌───────────┐     ┌────────────────┐     ┌─────────────┐ │
  │  │ Amazon S3 │────▶│ Knowledge Base │────▶│Bedrock Flows│ │
  │  │(Documents)│     │(Titan Embed V2 │     │(Visual RAG  │ │
  │  │           │     │ + Sync/Ingest) │     │ Pipeline)   │ │
  │  └───────────┘     └─────────┬──────┘     └───────┬─────┘ │
  │                       │                     │             │
  │                       ▼                     ▼             │
  │              ┌───────────────┐     ┌────────────┐         │
  │              │  OpenSearch   │     │ Foundation │         │
  │              │  Serverless   │     │Models      │         │
  │              │ (Vector Store)│     │(Claude +   │         │
  │              └───────────────┘     │ Titan Nova)│         │
  │                                    └────────────┘         │
  │                                                           │
  │  ┌──────────────────────────────────────────────────────┐ │
  │  │ Bedrock Agent — Code Interpreter (Python + Viz)      │ │
  │  └──────────────────────────────────────────────────────┘ │
  └───────────────────────────────────────────────────────────┘

---

📋 Prerequisites

Requirement	Details
AWS Account	With Amazon Bedrock access enabled
Model Access	Enable Claude 3.5 Haiku/Sonnet + Titan Text Embeddings V2 + Amazon Nova Micro in Model Access
IAM Permissions	Ability to create service roles, S3 buckets, OpenSearch collections
Source Documents	Included in this repo — AMZN-*-Shareholder-Letter.pdf

---

🎯 What You'll Build

Component	AWS Services	Purpose
Vector Store	OpenSearch Serverless	Store document embeddings for semantic search
Knowledge Base	Bedrock + S3 + Titan Embeddings V2	Ingest, chunk, and embed documents for RAG
RAG Pipeline	Bedrock Flows Visual Builder	Chain KB retrieval → prompt → model response
AI Agent	Bedrock Agent + Code Interpreter	Analyze data and generate Python visualizations

---

🚀 Section 1.1 — Create a Vector Store with OpenSearch Serverless

🧱 Step 1: Create an OpenSearch Vector Collection

📸 Expand for detailed steps

1. Navigate to Amazon OpenSearch Service in the AWS Console
2. On the left menu, go to Serverless → Click Dashboard
3. Click Get Started
4. Enter the following:

Setting	Value
Collection Name	bedrock-sample-rag
Collection Type	Vector Search
Deployment Type	✅ Enable Redundancy
Security	✅ Easy Create

5. Click Next, then Submit to create the collection

🧲 Step 2: Create the Vector Index

📸 Expand for detailed steps

1. After the collection is created, go to the Indexes section
2. Click Create Vector Index
3. Choose JSON mode
4. Use the following configuration:

{
  "settings": {
    "index.knn": "true",
    "number_of_shards": 1,
    "knn.algo_param.ef_search": 512,
    "number_of_replicas": 0
  },
  "mappings": {
    "properties": {
      "vector": {
        "type": "knn_vector",
        "dimension": 1024,
        "method": {
          "name": "hnsw",
          "engine": "faiss",
          "space_type": "l2"
        }
      },
      "text": {
        "type": "text"
      },
      "text-metadata": {
        "type": "text"
      }
    }
  }
}

5. Name your index: bedrock-sample-rag-index
6. Click Create

🧾 Important Metadata to Save

Property	Value
Index Name	bedrock-sample-rag-index
Vector Field Name	vector
Metadata Fields	text, text-metadata
Collection ARN	Save this from Console

---

📂 Section 1.2 — Upload Source Documents to Amazon S3

📸 Expand for detailed steps

1. Source documents (included in this repo):

   • AMZN-2019-Shareholder-Letter.pdf
   • AMZN-2020-Shareholder-Letter.pdf
   • AMZN-2021-Shareholder-Letter.pdf
   • AMZN-2022-Shareholder-Letter.pdf

2. Create an S3 bucket and upload all 4 documents

Setting	Recommendation
Bucket Name	bedrock-kb-<your-name>-lab (must be globally unique)
Region	Same region as your OpenSearch collection

---

🧠 Section 1.3 — Create a Bedrock Knowledge Base

📸 Expand for detailed steps

Create the Knowledge Base

1. Go to the Amazon Bedrock Console
2. In the left panel, click Builder Tools → Knowledge Bases
3. Click Create → choose Knowledge base with vector store
4. IAM: Select Create and use a new service role

Configure Data Source

Setting	Value
Source	Amazon S3
S3 Bucket	Select bucket from Section 1.2
Chunking Strategy	Fixed-size chunking
Max Tokens	512

🧬 Embeddings Model

1. Click Select Model.
2. Choose: Amazon Titan Text Embeddings V2.
3. Click Apply.

🧲 Vector Database Configuration

• Choose the previously created OpenSearch vector store.
• Ensure Vector engine for Amazon OpenSearch is selected.

Property	Value
Collection ARN	(Use ARN from Section 1.1)
Vector Index Name	bedrock-sample-rag-index
Vector Field Name	vector
Text Field Name	text
Metadata Field Name	text-metadata

Click Next to continue.

✅ Finalize

• In the Review and Create section, click Create Knowledge Base.
• After it’s created, go to the Data Sources tab.
• Select your source and click Sync to ingest documents into the knowledge base.

---

🧪 Section 1.4 — Test the Knowledge Base

📸 Expand for detailed steps

1. Go to your Knowledge Base in the Bedrock Console
2. Click Test Knowledge Base
3. Click Select Model → Choose Anthropic Claude 3.5 Haiku → Click Apply

Sample Query

What is Amazon doing in the field of generative AI?

4. Click Run and review the response
5. Click Show Details to inspect:
   • Extracted data chunks
   • Source metadata used for generation

---

🔄 Section 1.5 — Build a RAG Flow with Bedrock Flows Visual Builder

Build a visual flow that connects your Knowledge Base to a prompt template for dynamic question answering.

📸 Expand for detailed steps

Create the Flow

1. Go to the Amazon Bedrock Console → Click Flows.
2. Click Create Flow.
3. Enter Flow Name: langchain-kb-retriever
4. Under Service role name, select Create and use a new service role.
5. Click Create Flow.

🎯 Configure the Flow Builder

You'll now see 3 nodes: Flow input, Prompt, Flow output.

Add a Knowledge Base node from the menu.

Configure the Prompt Node

1. Click the Prompt Node → Select Model: Amazon Titan Nova Micro
2. Paste the following prompt template:

Human: You are a financial advisor AI system, and provide answers to questions using fact-based and statistical information when possible.  
Use the following pieces of information to provide a concise answer to the question enclosed in <question> tags.  
If you don't know the answer, just say that you don't know; don't try to make up an answer.

<question>
{{question}}
</question>

The response should be specific and use statistics or numbers when possible.

Context: {{context}}

A:

3. Under Prompt Settings → Change context input type to Array

Configure the Knowledge Base Node

1. Select the Knowledge Base from Section 1.3
2. Enable: ✅ Return retrieved results

Connect the Nodes

Flow input  ──▶  Prompt ({{question}})
Flow input  ──▶  Knowledge Base (retrieval query)
Knowledge Base  ──▶  Prompt ({{context}})
Prompt output  ──▶  Flow output

Click Save once connected.

Test the Flow

Use the Test Panel with this query:

What is Amazon doing in the field of generative AI?

Review the trace across nodes to confirm each step works correctly.

---

🤖 Section 2 — Create a Bedrock Agent with Code Interpreter

Build an AI agent that can write, execute, and debug Python code to analyze data and generate visualizations.

Setup

📸 Expand for detailed steps

1. Create the Agent

Setting	Value
Name	DataAnalysisAssistant
Description	AI Agent with Code Interpreter enabled for Data analysis
Model	Claude 3.5 Sonnet v2
Service Role	Create and use a new service role

2. Agent Instructions

You are an AI assistant specialized in data analysis and visualization.
You can write, run, and debug Python code to help users with their queries.
Always provide clear explanations of your process and results.

3. Enable Code Interpreter

• Expand Additional Settings
• Toggle ✅ Code Interpreter
• Ensure User Input is enabled

4. Save & Prepare

• Click Save → Click Prepare to activate the agent

Test the Agent

In the test panel, paste this prompt:

📋 Sample prompt with hotel satisfaction data

Using the customer satisfaction data provided below, create a bar graph
showing the average overall rating for each hotel (H001, H002, H003).
The graph should have the hotel IDs on the x-axis and the average overall
rating on the y-axis.

date,hotel_id,overall_rating,room_cleanliness,staff_service,amenities,value_for_money,location
2023-01-05,H001,4.2,4.5,4.0,3.8,4.1,4.5
2023-01-12,H002,3.8,4.0,3.5,3.9,3.7,4.2
2023-01-20,H003,4.5,4.7,4.6,4.3,4.2,4.8
2023-02-03,H001,3.9,4.2,3.7,3.6,3.8,4.4
2023-02-15,H002,4.1,4.3,4.0,4.1,3.9,4.3
2023-02-28,H003,4.3,4.5,4.4,4.2,4.0,4.6
2023-03-10,H001,4.0,4.3,3.8,3.7,3.9,4.5
2023-03-22,H002,3.7,3.9,3.6,3.8,3.5,4.1
2023-04-05,H003,4.4,4.6,4.5,4.1,4.1,4.7
2023-04-18,H001,4.1,4.4,3.9,3.8,4.0,4.6
2023-05-01,H002,3.9,4.1,3.8,4.0,3.6,4.2
2023-05-15,H003,4.6,4.8,4.7,4.4,4.3,4.9
2023-06-02,H001,4.3,4.6,4.1,4.0,4.2,4.7
2023-06-20,H002,4.0,4.2,3.9,4.1,3.8,4.3
2023-06-30,H003,4.5,4.7,4.6,4.3,4.2,4.8

Additional Test Prompts

Prompt	Tests
Create a heatmap showing correlation between satisfaction factors and overall rating	Data visualization
Calculate average scores for each satisfaction factor across all hotels	Data aggregation
Identify the hotel with the highest average rating and show its factor scores	Filtering + analysis
Create a line graph showing the trend of overall ratings over time	Time series

---

✅ Workshop Complete

You've built two end-to-end GenAI experiences using Amazon Bedrock, entirely through the console:

Experience	Components
RAG Pipeline	S3 → Titan Embeddings → OpenSearch → Knowledge Base → Flows → Claude/Titan
AI Agent	Bedrock Agent + Code Interpreter → Python analysis + visualizations

---

📚 Resources

Resource	Link
Amazon Bedrock Docs	docs.aws.amazon.com/bedrock
OpenSearch Service	docs.aws.amazon.com/opensearch-service
Knowledge Bases Guide	Knowledge Base Documentation
Bedrock Agents Guide	Agents Documentation
Original Workshop	catalog.workshops.aws

---

Built by Rodiel Lezcano · Solutions Architect