---

Multi-Agent Truth & Reasoning Checker

A modular, domain-specific validation system built using the LangChain framework. It uses external knowledge sources like Tavily Search, Wikipedia and DuckDuckGo, along with LLM-powered tools, to verify the correctness of statements in various domains — including math, factual knowledge, language, and code. This system does not generate new answers — it only checks and validates existing statements for accuracy and soundness.

---

Table of Contents

• Overview
• Project Structure
• Features
• Setup Instructions
• Usage
• Sample Inputs
• Dependencies
• Author

---

Overview

This system uses LangChain, an open framework for developing LLM-powered applications, to create a multi-agent architecture where each agent specializes in verifying different types of inputs.

The agents work collaboratively, ensuring that the main agent can intelligently delegate verification tasks, retrieve external information, and present fact-checked results with reasoning.

---

Project Structure

.
├── main_agent.py        # Main controller; routes statements to specific agents
├── router_agent.py      # Identifies domain using search tools
├── math_agent.py        # Validates math-related expressions and logic
├── english_agent.py     # Checks grammar, language correctness, and clarity
├── coding_agent.py      # Audits coding claims and language usage
├── factual_agent.py     # Uses Wikipedia & DuckDuckGo for fact-checking
├── requirements.txt     # All dependencies and LLM tooling
└── README.md            # This documentation

---

Features

• LangChain-Based Agent Architecture Designed using LangChain’s modular components, agents, and tool integrations.

• Domain Routing Agent Uses search-based heuristics to determine whether the claim is factual, mathematical, coding-related, or linguistic.

• Fact-Checker Queries Wikipedia and Tavily Search to validate real-world claims.

• Math Reasoning Checker Uses LLMMathChain for numerical expression verification.

• Code Logic Validator Audits basic programming knowledge across multiple languages and concepts.

• Language Assessment Agent Detects grammar errors and suggests fixes.

---

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/yourusername/multi-agent-verification-framework.git
   cd multi-agent-verification-framework

2. Install Dependencies:

   pip install -r requirements.txt

3. Add API Keys: Create a .env file and add:

   GOOGLE_API_KEY=your_google_api_key

---

Usage

Run the validation system:

python main_agent.py

• The system asks for user input (e.g., a claim or statement).
• It identifies the domain via router_agent.py.
• It then routes the input to the appropriate validation agent.
• The selected agent uses LLM + external tools to verify the truthfulness and explain the result.

---

Sample Inputs

Input Statement	Routed Agent	Result
"Python is a compiled language."	Factual Agent	Incorrect – it's interpreted
"He go to school everyday."	English Agent	Incorrect – grammar fixed
"2 + 2 × 3 = 12"	Math Agent	Incorrect – verified by order
"len('hello') returns 5 in Python."	Coding Agent	Correct – confirmed

---

Dependencies

Defined in requirements.txt. Includes:

• langchain
• langchain-google-genai
• langchain-community
• duckduckgo-search
• wikipedia-api
• python-dotenv

All built around the LangChain framework.

---

Author

Soumya Sourav Das

Portfolio | GitHub | LinkedIn

---