Project Overview🌟

The Deep Research Agent is a powerful AI-driven Multi-Agent System built with Composio and LangGraph, designed to conduct comprehensive research on user-specified topics within a chosen domain. It generates targeted research questions, performs in-depth analysis using AI-powered tools, and compiles findings into a professional, McKinsey-style HTML report, seamlessly saved to Google Docs. Leveraging the composio_langgraph library for tool integration and langchain_groq for language model interactions, this tool is perfect for researchers, analysts, or anyone seeking structured, high-quality insights. 🚀

With support for follow-up questions, it enables iterative refinement of research, making it a versatile solution for professional and academic use. 📊

Features ✨

• Input Flexibility 📝: Specify a research topic and domain (e.g., Health, Technology) via an intuitive Streamlit web interface.
• Automated Question Generation ❓: Generates three specific yes/no research questions tailored to the topic and domain.
• AI-Powered Research 🤖: Uses the Meta LLaMA model and COMPOSIO_SEARCH_TAVILY_SEARCH for real-time web searches to gather accurate data.
• Professional Reporting 📄: Compiles findings into a polished, HTML-formatted, McKinsey-style report.
• Google Docs Integration 📑: Automatically saves reports to Google Docs using GOOGLEDOCS_CREATE_DOCUMENT_MARKDOWN.
• Interactive Follow-Ups 🔄: Ask follow-up questions to refine or expand research results.
• State Management 🧮: Employs LangGraph for seamless workflow orchestration and memory management.

Project Structure 📂

├── notebook/deep_research.ipynb #notebook file
├── src/app.py                  # Main Streamlit application 
├── src/graph.py                # LangGraph workflow configuration 
├── src/state.py                # Graph state definition 
├── src/nodes/nodes.py          # Agent and tool nodes for the workflow 
├── src/tools/composio_tools.py # Composio toolset configuration 
├── src/tools/llm.py            # Language model setup 
├── src/prompts.py              # System prompt for the research agent 
├── .env                    # Environment variables 
├── requirements.txt #dependencies
├── License #license file
├── .gitignore #gitignore for env
├── . gitattributes #text= auto for normalization 
└── README.md               # Project documentation 📖

Tech Stack 🛠️

• Python 🐍: Core programming language.
• Streamlit 🌐: Powers the interactive web interface.
• LangGraph 🔗: Manages research workflow and state.
• LangChain (langchain_groq) 🤝: Interacts with the Meta LLaMA model.
• Composio 🔧: Enables web search (COMPOSIO_SEARCH_TAVILY_SEARCH) and Google Docs integration (GOOGLEDOCS_CREATE_DOCUMENT_MARKDOWN).
• dotenv 🔒: Securely manages API keys.
• MemorySaver 💾: Checkpoints and maintains research context across sessions.

Agents 🤖

The application utilizes a single agent named ResearchAgent, defined in nodes.py as the agent_node function. This agent is powered by the Meta LLaMA model (meta-llama/llama-4-scout-17b-16e-instruct) and is responsible for:

• Question Generation: Creates three targeted yes/no research questions based on user input using the language model.
• Research Execution: Processes research questions by invoking the language model and calling external tools to gather and analyze data.
• Report Compilation: Synthesizes findings into a professional, McKinsey-style HTML report for clarity and structure.
• Follow-Up Handling: Responds to user follow-up questions by leveraging prior context stored in the LangGraph memory.

The ResearchAgent operates within a stateful workflow, maintaining context across interactions using the MemorySaver for seamless research continuity.

Tools 🛠️

The application integrates the following tools via the ComposioToolSet from the composio_langgraph library, defined in composio_tools.py:

• COMPOSIO_SEARCH_TAVILY_SEARCH 🔍: A web search tool that fetches up-to-date information with configurable parameters (e.g., search_depth, max_results, include_images). Used to gather relevant data, statistics, and sources during research.

• GOOGLEDOCS_CREATE_DOCUMENT_MARKDOWN 📑: A tool that saves the compiled HTML report as a markdown document in Google Docs, enabling easy sharing and storage of research outputs.

• Report Writing with LLM A tool that uses LLM to write reports from getting answers from web search of given questions.

These tools are bound to the ResearchAgent using llm.bind_tools(tools) in nodes.py to enable function calling within the research workflow.

Example 📈

Input:

• Topic: AI-powered diagnostic tools
• Domain: Health

Output:

• Research Questions ❓:
  1. Are AI-powered diagnostic tools widely adopted in hospitals by 2025?
  2. Do AI diagnostic tools improve patient outcomes compared to traditional methods?
  3. Are there significant regulatory barriers to adopting AI diagnostic tools?
• Report 📄: A detailed HTML report with findings, saved to Google Docs.

Prerequisites ⚙️

• Python 3.10+
• Git
• LangChain
• Groq Cloud API key
• Optional: Streamlit (for UI)

Quick Start 🚀

git clone https://github.com/Zeeshier/deep-research-agent.git
cd deep-research-agent
pip install -r requirements.txt

# Set up your environment
touch .env
# Add your GROQ_API_KEY to the .env file

# Run the app
streamlit run app.py

Maintenance and Support 🛠️

This project is actively maintained. For questions, suggestions, or issues, please open an Issue or submit a Pull Request.

For direct inquiries, you can also reach me via:

• 📧 Email: zeeshanwarraich51@gmail.com
• 🌐 GitHub: @Zeeshier

Testing Methodology 🔍

• I used a curated benchmark of 20+ queries across domains (biology, AI, history).
• Each response was manually graded for relevance and correctness.
• Compared outputs to baseline models without RAG to measure improvement in contextual grounding.

License 📜

This project is licensed under the MIT License.
You are free to use, modify, and distribute this software with proper attribution.

For more details, please see the LICENSE file.