Repository files navigation

Bedrock Deep Research is a Streamlit-based application using Amazon Bedrock, LangGraph, and LangChain AWS libraries that automates article/report generation through AI-powered research, content writing, and image generation. It combines web research, structured content generation, and human feedback to produce comprehensive, well-researched articles with accompanying header images (generated byAmazon Bedrock Nova Canvas). This repo is inspired by LangChain'sDeep Researcher.

• Automated Research: Performs targeted web searches to gather relevant information
• Structured Content Generation: Creates cohesive article outlines and detailed section content
• Interactive Feedback Loop: Incorporates human feedback to refine article outlines
• AI-Generated Imagery: Produces relevant header images for visual appeal

bedrock_deep_research/├── bedrock_deep_research.py          # Main Streamlit application entry point├── bedrock_deep_research/│   ├── config.py             # Configuration settings and parameters│   ├── graph.py              # Core workflow orchestration using LangGraph│   ├── model.py              # Data models for articles and sections│   ├── nodes/                # Individual workflow components│   │   ├── article_head_image_generator.py    # Header image generation│   │   ├── article_outline_generator.py       # Article outline creation│   │   ├── section_writer.py                  # Section content generation│   │   └── [other node files]                 # Additional workflow components│   ├── utils.py              # Utility functions│   └── web_search.py         # Web research integration using Tavily API├── poetry.lock               # Poetry dependency lock file└── pyproject.toml           # Project configuration and dependencies

The application follows a sequential workflow from topic input to final article generation, with feedback loops for refinement.

Key Components of the graph:

1. Initial Researcher: It performs initial web searches to gather context
2. Article Outline Generator: creates structured outline using research data
   • Human Feedback Provider: This incorporates human feedback for the outline
3. Section Writer: A subgraph that generates content after web research.
4. Compilation: combines all elements into a cohesive article
5. Final Section Generation: Generate the overview and the last paragraph based on the other sections.
6. Header Image Generator: creates relevant header image

The setup is meant to be used locally withAWS authentication, as well as within Amazon Sagemaker: either inJupyterLab orCode Editor instance.

Note: Current setup is assumingus-east-1 region (as defined inenv.tmp file).

• Python 3.12 (to install, visit this link:https://www.python.org/downloads/).Check your python version usingpython --version.If your global python isn't set as3.12, follow the steps here:https://python-poetry.org/docs/managing-environments/)
• Poetry for dependency management
• Make sure you haveenabled model access viaAWS Bedrock access inus-east-1 region. You can find the supported models provided inSUPPORTED_MODELS variable in./bedrock_deep_research/config.py.
• Tavily API key for web research capabilities.

# Clone the repositorygit clone https://github.com/aws-samples/sample-bedrock-deep-researcher.gitcd sample-bedrock-deep-researcher# Activate the virtual environmentpoetry shell# Install dependencies using Poetrypoetry install

Go tohttps://app.tavily.com/home and create a free API KEY. Copy the API Key and paste it into theenv.tmp file.

Copy the environment variables into the local environment.

cp env.tmp .env

• Start the Streamlit application:

streamlit run bedrock_deep_research.py

Step	Description	Visual Reference
1. Enter Article Details	• Enter your article topic in the main input field • Add specific writing guidelines in the text area provided • Adjust search parameters using the configuration panel • Click the "Generate Outline" button to start the process
2. Review and Refine the Outline	• Review the AI-generated article outline • Provide specific feedback in the feedback field to improve the structure • Use the editing tools to make direct modifications if needed • Click "Accept Outline" when you're satisfied with the structure
3. Generate the Complete Article	• Review the fully researched and written article with its custom header image • Use the formatting tools to make any final adjustments • Click "Copy to Clipboard" to export your article • Or select "New Article" to start the process again

Writing Guideline samples:You could include instructions like:

- Strict 150-200 word limit- Start with your most important insight in **bold**- Include code examples where relevant- Focus on practical implementation

Web Research Configuration:

number_of_queries=2# Number of search queries per sectionmax_search_depth=2# Maximum research iterations per section

Debug Mode:

# Enable debug loggingexport LOG_LEVEL=DEBUGstreamlit run bedrock_deep_research.py

Contributions are welcome! Please open an issue or submit a pull request if you have any improvements or bug fixes. Read CONTRIBUTING.md for more details.

This library is licensed under the MIT-0 License. See the LICENSE file.