10-multi-modal-email-assistant-agent

✉️ 🔍 Email Assistant with RAG and Image Generation

Author: Julia Hu

This project presents a multi-agent Email Assistant that leverages retrieval-augmented generation (RAG), image synthesis, and web data enrichment to generate high-quality, multi-modal email content. Built using the Strands Agents SDK, the assistant coordinates a suite of specialized agents—each responsible for tasks like querying knowledge bases, generating contextual images, and refining email content.

The system architecture integrates Amazon Bedrock for both text and image generation, a Knowledge Base for enterprise data retrieval, and optional web research capabilities to supplement real-time context. The final output is a well-structured, polished email that includes relevant knowledge snippets and visually compelling elements, all orchestrated through a single plain-language request.

1. What is this?

A comprehensive email assistant that leverages knowledge base retrieval and image generation capabilities.

Function	Agent/Tool	Data Source	Output
Orchestration	email_assistant	Coordinates tools & writes final email	Structured Markdown / HTML
Knowledge Retrieval	retrieve_from_kb	Amazon Bedrock Knowledge Base	Relevant context from audio/text
Image Creation	generate_image_nova	Amazon Bedrock Nova Canvas	Custom images for email content
Web Research	http_request	Public web	Supplementary information
Content Editing	editor	Amazon Nova	Polished email content

2. Flow overview 🚦

1. User ➜ email_assistant with a plain-language email request
2. Assistant analyzes request and determines needed resources:
   • Knowledge context? → calls retrieve_from_kb to query Knowledge Base
   • Visual content? → calls generate_image_nova to create relevant images
   • Additional facts? → uses http_request for web research
3. Assistant synthesizes all gathered information
4. Final polished email returned to the user with embedded images and context

3. Email format 📋

The assistant creates professional emails with:

• Proper structure (greeting, body, closing)
• Relevant knowledge base context integrated naturally
• Custom-generated images when appropriate
• Web research findings incorporated seamlessly
• Professional tone and formatting

4. Quick start 🛠️

Prerequisites

• Python 3.10+
• AWS CLI v2 configured
• Access to Amazon Bedrock (Nova in us-east-1 region and Knowledge Base in your preferred region)

Step 1: Set up Knowledge Base

1. Run the notebooks in Multi-modal-data-ingest/audio-video-rag/ folder to process audio files
2. Create a Bedrock Knowledge Base and note the KB ID

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Run the Email Assistant

python email_assistant.py --kb-id YOUR_KB_ID --region replace_your_region

Optional: Run individual agents

Run the KB RAG agent

python kb_rag.py --kb-id YOUR_KB_ID

Run the image generation agent

python image_generation_agent.py

5. AWS Architecture 🏗️

Component Type	AWS Service	Purpose
Data Pipeline	Bedrock Data Automation	Processes audio files into embeddings
Storage Layer	S3	Stores documents, images, and processed data
Search Layer	Amazon Bedrock Knowledge Base	Vector/keyword index for knowledge retrieval
AI Services	Amazon Nova	Foundation model for text generation and image creation
Compute	Amazon SageMaker	Notebook environment for development

6. Troubleshooting 🐞

Symptom	Likely Cause	Fix
NoCredentialsError	AWS credentials not exported	Run aws configure --profile <profile>
Knowledge Base Error	KB ID incorrect or in wrong region	Verify KB ID and region match where KB was created
Image Generation Fails	Token limit exceeded	Reset conversation after each interaction
ImportError Strands SDK	Missing Strands SDK	Check Python path includes SDK locations
ModuleNotFoundError	Incorrect import structure	Use absolute imports when running as script

7. Usage Examples

Basic Email Request

Email Request> Write an email to my team about our Q3 financial results

Email with Knowledge Base Context

Email Request> Write an email summarizing the key points from our last earnings call

Email with Custom Image

Email Request> Write an email announcing our new product launch with an image of a futuristic device

8. Advanced Configuration

You can customize the Email Assistant by modifying:

• The system prompt in create_email_assistant() function
• The default region and KB ID in the script
• The tools available to the assistant

For production use, consider:

• Setting up persistent storage for generated images
• Implementing authentication for sensitive knowledge bases
• Creating a web interface using Streamlit or Flask

9. Cleanup

After you're done using the Email Assistant, you should clean up the resources to avoid incurring unnecessary costs.

Option 1: Clean up in the notebook

If you created the knowledge base using the notebook in Multi-modal-data-ingest/audio-video-rag/, you can run the cleanup cell at the end of the notebook:

# Delete Knowledge Base
knowledge_base.delete_kb(delete_s3_bucket=True, delete_iam_roles_and_policies=True)