AI vs AI
Multi-Agent Architecture for improved reasoning.
Overview:
I developed an app that facilitates debates between two AI models, allowing users to observe their arguments unfold. The app leverages the same technology stack as the Shana's Sourdough chatbot, reusing much of the infrastructure to create a platform where AI models engage in structured debates on user-specified topics.
Try the Debate Machine out here!
Technologies Used:
-
LangChain Framework: Used to set up AI agents as expert debaters, allowing for flexible prompting and argument generation.
-
OpenAI 3.5 Turbo Model: Powers the conversational abilities of the AI debaters, ensuring coherent and contextually relevant arguments.
-
Python: Utilized for backend processing and data management.
-
React and CSS: Implemented for the frontend UI design, providing an intuitive user experience.
-
Vercel: Chosen for hosting the frontend, offering seamless deployment and serverless architecture.
-
AWS Lambda Functions and DynamoDB: Utilized for serverless backend development, enabling efficient handling of debate sessions and message persistence.
The Project:
Concept and Initial Setup:
Building on the success of the Shana's Sourdough chatbot, I aimed to explore the concept of AI models debating each other. By leveraging the LangChain framework, I set up AI agents with custom prompts, positioning them as expert debaters on opposing sides of a given topic.
Development of Debate Mechanics:
The app allows users to specify a debate topic or choose from premade topics, with the option to edit or expand on the topic. Users can also customize the prompts for the Affirmative and Negative models, though default prompts are provided to streamline the process. The debate mechanics were designed to alternate between the models, simulating a real debate format.
UI Design and Deployment:
I designed the frontend using React and CSS, ensuring a user-friendly interface that allows users to easily set up and follow the debates. Vercel was utilized for hosting, providing a straightforward deployment process and serverless infrastructure.
Backend Implementation:
The backend was developed using AWS Lambda functions to handle the debate logic and persist messages in a DynamoDB database. This serverless approach made for a hassle free set up to create a reliable backend for my application.
Achievements:
By developing this debate app, I demonstrated the versatility of AI models and the potential for interactive applications. This project was leveraged by my ability to reuse and adapt my existing infrastructure in a new way. Hopefully this sparks some fun ideas for what you would like to see debated or what's possible with multi agent architecture.