Hacktober 2026 : Build for a friend.
Meet RoundTable AI — a voice-first multi-agent room where AI perspectives debate your decisions instead of simply giving you an answer.
What I Built
When a friend has an important decision to make, sometimes they don't need another person to simply tell them what to do. They need different perspectives.
That's why I built RoundTable AI—a voice-first, multi-agent decision-making platform where different AI perspectives come together in a virtual meeting room and debate a decision.
The idea is simple:
Don't ask AI for an answer. Build a room that argues about it.
The user provides a decision or problem, and RoundTable creates relevant AI participants with different roles, goals, expertise, and opinions.
For example, the room can include an Investor, Engineer, Customer, Competitor, Devil's Advocate, and Moderator.
The agents don't simply give separate answers. They challenge each other's arguments, question assumptions, and work toward a structured Room Verdict.
The user also stays part of the conversation and can interrupt, challenge the room, ask questions, or add another perspective.
I built RoundTable AI for a friend who often has to make important decisions and could benefit from having multiple viewpoints available before making a choice.
Code
The project is open source:
GitHub: RoundTable AI on GitHub
How I Built It
RoundTable AI is built around three main parts.
🧠 Gemma — The AI Reasoning Layer
The reasoning layer uses Gemma, an open-weight AI model, to power the agents, their perspectives, arguments, challenges, and final synthesis.
The architecture is designed to be model-agnostic, allowing us to experiment with different open models and local inference.
🎙️ ElevenLabs
I use ElevenLabs for the voice layer.
Each AI participant can have a distinct voice, making the experience feel more like an actual meeting instead of a normal chatbot conversation.
⚡ Real-Time Application
The application is built with:
Next.js
React
TypeScript
Tailwind CSS
Node.js
Fastify
Socket.IO
The backend manages the meeting room, participants, conversation, turns, challenges, and final verdict.
An Argument Ledger keeps track of important claims and disagreements so the agents can respond to what is actually being discussed.
Why Does Open Innovation Matter?
Open-weight AI isn't just a checkbox for this project.
RoundTable is designed around experimenting with how different AI agents reason, challenge each other, and reach a decision.
Using open-weight AI gives us the freedom to:
Swap models
Experiment with different models for different roles
Modify agent behavior
Explore local inference
Reduce dependency on a single AI provider
Potentially keep sensitive decisions local
The reasoning model can evolve without rebuilding the entire application.
That's important because RoundTable is not about one particular AI model. It's about creating a system where different AI perspectives can interact.
Open-weight AI provides the reasoning.
RoundTable provides the room.
ElevenLabs provides the voice.
My Agent Session
I used DevRelay during development to help with architecture, implementation planning, code generation, and reviewing the multi-agent system.
Prize Categories
🎙️ ElevenLabs — For giving each AI participant a distinct voice and creating a voice-first meeting experience.
🧠 Gemma — For powering the open-weight AI reasoning and multi-agent debate.
Final Thoughts
We already have AI assistants that answer questions.
I wanted to build something different.
Instead of giving you one AI answer, RoundTable gives you a room full of perspectives.
Sometimes the best AI isn't the one that agrees with you.
It's the one that makes you think twice.
RoundTable AI — Don't ask AI for an answer. Build a room that argues about it.
Top comments (0)