Agents-MCP-Hackathon/MCP_ClaimVerificationSystem

1---2title: Claim Verfication System using Kognie API3emoji: 🚀4colorFrom: yellow5colorTo: blue6sdk: gradio7sdk_version: 5.33.18app_file: app.py9pinned: false10license: apache-2.011short_description: Multi Agentic Claim Verification System - Track 112tags: ["mcp-server-track"]13---14 15 16# 🔍 Multi-Agent Claim Verification System using Kognie API17 18An intelligent, multi-agent MCP server designed to verify claims using diverse AI models and real-time web research. This system combines the power of multiple language models with web search capabilities to provide comprehensive fact-checking and evidence analysis.19 20Visit : https://kognie.com/api to create a Kognie API key to gain access to multiple LLMs with one single account.21 22## 📽️ Demo video23We have used Github Copilot as the MCP client for our MCP server.24 25**URL** : https://drive.google.com/file/d/1-vczaQAsA9-wxwzlg91fCrQT18JYkmN6/view?usp=sharing26 27**MCP client model** : Claude 3.7 Sonnet28 29## 🎯 Purpose30 31In an era of information overload and misinformation, this system serves as a robust fact-checking tool that:32 33- **Verifies claims** using multiple AI perspectives34- **Gathers real-time evidence** from web sources35- **Provides balanced analysis** with supporting and contradicting evidence36- **Makes informed decisions** based on comprehensive data analysis37- **Presents results** in an intuitive, interactive web interface38 39## 🏗️ System Architecture40 41The system employs a **hierarchical multi-agent architecture** with specialized roles:42 43### 🤖 Agent Specifications44 45 46#### 1. **MultiLLM Verifier Agent**47- **Model**: Claude-3.5-Sonnet (Anthropic)48- **Role**: Cross-model evidence analysis49- **Responsibilities**:50  - Coordinates multiple LLM perspectives51  - Runs parallel analysis across different AI models52  - Provides diverse viewpoints on claims53- **Internal process**:54  The system leverages **three distinct AI models** for comprehensive analysis:55 56  | Model | Provider | Strengths |57  |-------|----------|-----------|58  | **GPT-4o-mini** | Kognie API | Fast reasoning, general knowledge |59  | **Gemini-2.0-Flash** | Kognie API | Multimodal capabilities, recent training |60  | **Open-Mistral-Nemo** | Kognie API | European perspective, specialized domains |61 62  ##### Parallel Processing Benefits63  - **Diverse Perspectives**: Each model brings unique training and biases64  - **Cross-Validation**: Multiple viewpoints reduce single-model limitations65  - **Speed**: Asynchronous processing ensures rapid results66  - **Robustness**: System continues functioning even if one model fails67 68 69#### 2. **Web Evidence Retriever Agent**70- **Model**: Claude-3.5-Sonnet (Anthropic)71- **Role**: Real-time information gathering72- **Responsibilities**:73  - Searches current web sources74  - Retrieves up-to-date information75  - Provides context-aware evidence76- **Internal process**:77  ##### Real-Time Evidence Gathering78  - **Bing Search API** integration for current information79  - **News source prioritization** for recent developments80  - **Automated query generation** based on claim analysis81  - **Evidence categorization** (supporting vs. contradicting)82 83  ##### Search Strategy84  1. **Query Optimization**: Transforms claims into effective search terms85  2. **Source Diversification**: Gathers information from multiple web sources86  3. **Recency Prioritization**: Focuses on current and relevant information87  4. **Result Synthesis**: Analyzes and structures findings88 89 90#### 3. **Boss Agent** (Coordinator)91- **Model**: GPT-4o (OpenAI)92- **Role**: Final decision maker and coordinator93- **Responsibilities**:94  - Orchestrates other agents95  - Synthesizes evidence from multiple sources96  - Makes final verification decisions97  - Formats results in HTML for presentation98 99## 💻 User Interface100 101### Interactive Web Interface (Gradio)102- **Chat-based interaction** for natural claim submission103- **Real-time processing** with progress indicators104- **Collapsible analysis sections** for detailed evidence review105- **Color-coded results** (Green for TRUE, Red for FALSE)106- **Responsive design** for various devices107 108### Key Features109- **Instant verification** results110- **Detailed evidence breakdown** from each agent111- **Interactive expandable sections** for in-depth analysis112- **Clean, professional presentation** of complex data113 114 115## 🚀 Getting Started116 117### Prerequisites118```bash119pip install gradio llama-index python-dotenv asyncio120```121 122### Environment Variables123Create a `.env` file with the following:124```env125KOGNIE_BASE_URL=your_kognie_base_url126KOGNIE_API_KEY=your_kognie_api_key127BING_SUBSCRIPTION_KEY=your_bing_api_key128BING_SEARCH_URL=your_bing_search_url129ANTHROPIC_API_KEY=your_anthropic_api_key130OPENAI_API_KEY=your_openai_api_key131MISTRAL_API_KEY=your_mistral_api_key132```133 134### Running the Application135```bash136gradio app.py137```138 139The system will launch a web interface accessible through your browser.140 141## 🎯 Use Cases142 143### Perfect For:144- **Fact-checking news claims**145- **Academic research verification**146- **Social media post validation**147- **Business claim analysis**148- **Educational fact verification**149- **Journalism and reporting**150 151### Example Claims:152- "Company X reported record profits in Q4 2024"153- "New scientific study proves Y causes Z"154- "Political candidate made statement about policy"155- "Sports team won championship in specific year"156 157## 🔮 Technical Advantages158 159### 1. **Asynchronous Processing**160- Non-blocking operations for faster results161- Concurrent agent execution162- Responsive user interface163 164### 2. **Error Resilience**165- Graceful handling of API failures166- Fallback mechanisms for each component167- Comprehensive error logging168 169### 3. **Scalable Architecture**170- Easy addition of new AI models171- Modular agent design172- Configurable processing parameters173 174### 4. **Evidence Transparency**175- Complete audit trail of analysis176- Source attribution for all evidence177- Detailed reasoning for decisions178 179## 🛡️ Quality Assurance180 181### Multi-Layer Verification1821. **Cross-Model Validation**: Multiple AI perspectives1832. **Real-Time Research**: Current information priority1843. **Evidence Weighting**: Web sources prioritized for recent events1854. **Transparent Reasoning**: Complete decision audit trail186 187### Bias Mitigation188- **Model Diversity**: Different training approaches and datasets189- **Source Variety**: Multiple web sources and perspectives190- **Temporal Awareness**: Prioritizes recent information191- **Evidence Balance**: Seeks both supporting and contradicting evidence192 193## 🔧 Customization Options194 195The system is designed for easy customization:196- **Add new AI models** to the MultiLLM verifier197- **Integrate additional search engines** beyond Bing198- **Modify decision-making logic** in the Boss Agent199- **Customize UI themes** and presentation styles200- **Adjust evidence weighting** algorithms201 202## 🤝 Contributing203 204This system represents a foundation for intelligent claim verification. Areas for enhancement include:205- Additional AI model integrations206- Advanced evidence scoring algorithms207- Specialized domain knowledge bases208- Multi-language support209- API endpoint creation210 211 212