.png)
WHISTLE (Weighted Hybrid Intelligent System for Trust, Logic & Evidence) is a multi-agent AI verification system designed to improve trust in AI-generated answers.
Instead of relying on a single AI model, WHISTLE compares responses from multiple AI agents and applies consensus logic and confidence scoring to produce a more reliable final answer.
This project is developed as an academic and hackathon prototype exploring how AI outputs can be verified before being trusted.
CURRENT STATUS: PROTOTYPE // DEMONSTRATION MODE The system currently runs in a simulated environment to demonstrate UI, workflow orchestration, and multi-agent verification logic.
Modern AI systems provide fast responses but may generate incorrect or misleading information. WHISTLE addresses this challenge by introducing a verification layer where multiple AI agents independently analyze the same input.
The system then:
- Compares outputs
- Identifies agreement and conflicts
- Produces a verified answer
- Assigns a confidence score
The goal is to make AI usage more transparent, explainable, and trustworthy.
WHISTLE is built around a multi-agent consensus workflow:
-
User Query Input A question or task is submitted.
-
Parallel AI Analysis Multiple AI agents process the input independently.
-
Consensus & Weighting Responses are compared using agreement-based logic.
-
Verified Output A final answer is generated with a confidence score.
-
(Optional) Trust Layer Verified results can be recorded for audit or transparency purposes.
Uses multiple AI agents instead of a single model to reduce reliance on one perspective.
Identifies common conclusions between AI outputs and reduces the influence of outliers.
Each verified answer includes a confidence level based on response agreement.
Users can see why a particular answer was selected.
Designed to integrate with workflow systems (such as Chainlink CRE) for verifiable execution in hackathon environments.
- React (Vite)
- TypeScript
- Tailwind CSS
- JavaScript-based mock AI responses
- Consensus and scoring logic implemented on the client side for demonstration
- Chainlink CRE for workflow orchestration
- Smart contract interaction for verifiable result recording
WHISTLE/
├── src/
│ ├── App.tsx # Main application router
│ ├── pages/
│ │ ├── Home.tsx # Product overview
│ │ ├── About.tsx # Project background and team
│ │ ├── Demo.tsx # AI verification demo interface
│ │ ├── Blog.tsx # Project articles and updates
│ │ └── Login.tsx # Future dashboard access (UI only)
│ ├── components/
│ │ ├── ChatInterface.tsx # Multi-agent verification simulation
│ │ ├── FeatureCards.tsx
│ │ ├── WorkflowSteps.tsx
│ │ └── Navbar.tsx
└── README.md
For detailed setup guides, troubleshooting, and deployment instructions, please verify the Setup Documentation.
- Node.js (v18 or later)
- npm or yarn
git clone https://github.com/your-repo/whistle.git
cd whistle
npm install
npm run devVisit: http://localhost:5173
- Navigate to the Demo page.
- Enter a question in the input field.
- Click Run Verification.
- The system will simulate:
- Multiple AI responses
- Consensus analysis
- Final verified answer
- Confidence score
Note: Current AI responses are simulated for demonstration purposes.
- Bathrinath
- Priyadharshan
- Dharshan
- Vasudevan
- Real AI API integration
- Dynamic agent weighting
- Domain-specific verification models
- On-chain storage of verified outputs
- Enhanced confidence scoring methods
WHISTLE is an academic and hackathon prototype. It is intended for research and demonstration purposes only and does not provide professional or automated decision-making services.
MIT License