Inspiration
We were inspired to create this project after seeing how traditional pros and cons lists often fall short of delivering the depth needed for truly informed decision-making. We wanted to build something that goes beyond simple lists—something that mimics a real internal debate, where two opposing perspectives battle it out to uncover the nuances of an issue.
What it does
DebateHub enables two AI agents to engage in a structured debate, with one arguing for and the other against a topic. They go back and forth over a set number of turns, responding to each other’s points while introducing new arguments. A third AI agent steps in as the judge, evaluating the debate based on criteria like argument strength, evidence use, and rebuttal quality. At the end, it provides users with a summary and declares a winner based on the scores.
How we built it
We used the OpenAI python library, along with the Gemini 1.5 Flash model through the OpenRouter API to create three distinct AI agents: one pro, one con, and a judge. The agents are initialized with specific prompts to guide their behaviors and maintain their biases. The debate itself runs through a loop where each agent takes turns presenting and countering arguments. The judge agent then analyzes the entire conversation, scores both sides using a structured rubric, and compiles a balanced report.
Challenges we ran into
One big challenge was making sure the agents stayed true to their assigned biases while addressing their opponent’s arguments and introducing fresh points. Another was ensuring that both agents interacted with each other's responses effectively, rather than repeating or ignoring key points. The toughest part, though, was creating an unbiased judge agent that could objectively evaluate the debate without tilting toward one side. It took a lot of adjustments to get the evaluation process to feel fair and balanced, and there is still more work to be done.
Accomplishments that we're proud of
We’re proud that we were able to build a system where AI agents don’t just talk past each other but actually engage in a meaningful debate that helps users see multiple sides of an issue. We’re also proud of the judge agent we developed—it’s capable of weighing the debate fairly and producing a summary that’s easy for users to understand.
What we learned
We learned how to design AI agents that can stay true to their roles and biases while engaging in structured argumentation. We also learned a lot about the challenges of creating an evaluation process that’s both thorough and unbiased. This project taught us the importance of balancing depth with simplicity to make the AI’s output both insightful and user-friendly.
What's next for DebateHub
Next, we want to add research capabilities (search grounding) so that the agents can pull in external evidence and citations to back up their arguments. We also plan to fine-tune the scoring system to better reflect nuances like logical consistency and source reliability. Ultimately, we hope to make DebateHub even more interactive and insightful for users seeking balanced perspectives.
Log in or sign up for Devpost to join the conversation.