A sophisticated research automation tool that leverages LangChain and LangGraph to conduct in-depth research on any topic. The system uses multiple AI agents working together to gather information, analyze data, and generate comprehensive research reports.

• Python 3.12 or higher
• Poetry for dependency management
• API keys for:
  • OpenAI
  • Anthropic
  • Google AI
  • Tavily (Mandatory)

The API key of the LLM provider is optional based on which model you wish to use.

1. Clone the repository:

git clone https://github.com/psykick-21/deep-research.git
cd deep-research

2. Create a virtual python environment in this repo

conda create -p venv python=3.12 -y

Any other method can also be used to create python environment.

3. Activate python environment

conda activate ./venv

4. Install poetry in the environment if you are not using global poetry config

pip install poetry

5. Install dependencies using Poetry:

poetry install

6. Create a .env file in the project root with your API keys:

OPENAI_API_KEY=your_openai_key
ANTHROPIC_API_KEY=your_anthropic_key
GOOGLE_API_KEY=your_google_key
TAVILY_API_KEY=your_tavily_key

Change the topic and outline based on your requirements in the main.py file.

Run the research workflow:

python main.py

Alternatively, you can use main.ipynb notebook for a more interactive experience.

• Automated research workflow using multiple specialized AI agents
• Deep web search capabilities using Tavily
• Structured report generation with customizable sections
• Human-in-the-loop feedback system
• Configurable research parameters (depth, temperature, etc.)
• Detailed logging and progress tracking

The research process follows a sophisticated workflow:

1. Initial Planning Phase

   • User provides the research topic and outline
   • Report Structure Generator creates a detailed outline
   • Human feedback can be incorporated to refine the structure

2. Section-wise Research

   • The outline is broken down into multiple sections
   • Each section undergoes a detailed research process:
     • Utilizes existing LLM knowledge base
     • Performs targeted web searches
     • Accumulates and processes search results
     • Reflects on gathered information
     • Generates additional search queries based on reflection
     • Synthesizes final section content

3. Final Curation

   • All sections are combined and refined
   • Final report is generated with proper formatting and structure

The system employs a recursive improvement loop where each section's research can trigger multiple rounds of:

• Web searches
• Knowledge base consultation
• Result accumulation
• Reflection
• Additional query generation

This ensures comprehensive coverage of the topic while maintaining coherence and depth in the research.

• deep_research/: Main package directory

  • nodes.py: Contains the implementation of various research agents
  • graph.py: Defines the LangGraph workflow
  • prompts.py: Contains the prompts used by different agents
  • struct.py: Defines the data structures used in the workflow
  • state.py: Manages the state of the research process
  • utils.py: Utility functions
  • configuration.py: Configuration settings

• logs/: Contains detailed logs of the research process

• reports/: Stores the generated research reports

• main.py: Entry point of the application

You can customize the research process by modifying the following parameters in main.py:

• max_queries: Maximum number of search queries to perform
• search_depth: Depth of the research
• num_reflections: Number of reflection cycles
• temperature: Controls the creativity of the AI responses

This project is licensed under the terms included in the LICENSE file.

Gaurav Sharma ([email protected])