ResearchMind

ResearchMind is a multi-agent AI research system that automates web research, content extraction, report generation, and report evaluation.

The system uses specialized AI agents and LangChain pipelines to transform a research topic into a structured research report with critic feedback.

Live Demo

Launch ResearchMind

Overview

ResearchMind follows a four-stage research workflow:

1. Search Agent - Searches the web for recent and relevant information.
2. Reader Agent - Selects and scrapes relevant web resources for deeper context.
3. Writer Chain - Generates a structured research report from the collected information.
4. Critic Chain - Reviews the generated report, assigns a score, and provides strengths and areas for improvement.

The application provides an interactive Streamlit interface for running the complete research pipeline.

Architecture

User Research Topic
        |
        v
+-------------------+
|   Search Agent    |
|   Tavily Search   |
+-------------------+
        |
        v
+-------------------+
|   Reader Agent    |
|   Web Scraping    |
+-------------------+
        |
        v
+-------------------+
|   Writer Chain    |
|   Report Creation |
+-------------------+
        |
        v
+-------------------+
|   Critic Chain    |
| Review and Score  |
+-------------------+
        |
        v
Final Research Report
and Critic Feedback

Features

• Multi-agent AI research workflow
• Real-time web search using Tavily
• Web content extraction with BeautifulSoup
• Agent-based tool calling with LangChain
• Mistral LLM integration
• LCEL-based writer and critic chains
• Structured research report generation
• Automated report evaluation and scoring
• Streamlit web interface
• Pipeline progress visualization
• Raw search and scraped-content inspection
• Markdown report download

Agent Workflow

Search Agent

The Search Agent uses a Tavily-powered web search tool to gather recent and reliable information related to the research topic.

Reader Agent

The Reader Agent uses a scraping tool to extract clean text from a relevant web resource. Scripts, styles, navigation elements, and footer content are removed before text extraction.

Writer Chain

The Writer Chain combines search results and scraped content to generate a professional research report containing:

• Introduction
• Key Findings
• Conclusion
• Sources

Critic Chain

The Critic Chain evaluates the generated report and returns:

• Score out of 10
• Strengths
• Areas to Improve
• One-line verdict

Tech Stack

• Python
• LangChain
• LangChain Agents
• LangChain Expression Language (LCEL)
• Mistral AI
• Tavily Search API
• Streamlit
• BeautifulSoup
• Requests
• python-dotenv

Project Structure

MULTIAGENT_SYSTEM/
|
|-- app.py
|-- agents.py
|-- tools.py
|-- pipeline.py
|-- requirements.txt
|-- .gitignore
`-- README.md

File Description

---

File Purpose

---

app.py Streamlit user interface and interactive pipeline execution

agents.py Search Agent, Reader Agent, Writer Chain, and Critic Chain

tools.py Tavily web search and URL scraping tools

pipeline.py Command-line orchestration of the complete research pipeline

requirements.txt Python project dependencies

---

Installation

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/ResearchMind.git
cd ResearchMind

2. Create a virtual environment

python -m venv venv

Activate it on Windows:

venv\Scripts\activate

Activate it on macOS or Linux:

source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Create a .env file in the project root:

MISTRAL_API_KEY=your_mistral_api_key
TAVILY_API_KEY=your_tavily_api_key

Never commit the .env file or expose API keys publicly.

Run the Application

Start the Streamlit application:

python -m streamlit run app.py

The application will open in your browser. Enter a research topic and run the research pipeline.

Run the Command-Line Pipeline

You can also execute the research workflow from the terminal:

python pipeline.py

Enter a research topic when prompted.

Example Workflow

Research Topic:
LLM agents in 2025

Search Agent
    |
    v
Collects recent web results

Reader Agent
    |
    v
Extracts detailed content

Writer Chain
    |
    v
Generates structured report

Critic Chain
    |
    v
Scores and reviews report

Key Concepts Demonstrated

This project demonstrates:

• Multi-agent AI systems
• Tool calling
• Agent specialization
• Web search integration
• Web scraping
• LLM orchestration
• LangChain agents
• LCEL pipelines
• Prompt engineering
• State-based pipeline execution
• AI-generated content evaluation

Author

Mehar Arora