VeriVox

HeroShot
Game Demo
Main Lobby
Logo

Inspiration

We were inspired by the shortcomings of traditional text-based debate and the untapped potential of voice-assisted interaction. We wanted to combine symbolic logic checks with AI-driven voice to create a more immersive, structured, and engaging debate and learning experience. Our goal was to empower users with stronger reasoning skills in a fun, community-driven environment.

What it does

VeriVox is a voice-oriented debate and learning platform that:

Integrates ElevenLabs voice APIs for real-time AI speech interaction.
Uses audio reflective particles to visualize user engagement and states like recording, speaking, or standby.
Incorporates a symbolic logic engine that validates the soundness of user statements (something typical LLMs struggle with).
Teaches five fundamental logic symbols (And, Or, Not, If…then, If and only if) through an interactive tutorial.
Offers an online multiplayer mode with achievements, letting users debate, play, and hone their logic skills together.

How we built it

Voice Integration
- Leveraged ElevenLabs for natural-sounding AI voice and real-time speech output.
- Built a custom voice pipeline with web audio APIs to capture user input and provide swift feedback.
Symbolic Logic System
- Developed a custom logic validator that parses user statements and checks their internal consistency.
- Connected it to LLMs (GPT-4o) for natural language processing—while the formal logic engine does the “heavy lifting” on validity.
Audio Reflective Particles
- Implemented a visual layer in the frontend (React + WebGL or Canvas) that reacts dynamically to user voice states (e.g., “Recording,” “Listening,” “Idle”).
Online Game & Achievements
- Created a multiplayer environment using Node.js + websockets, allowing real-time debate sessions.
- Implemented an achievement system in the database to track user progress (e.g., number of debates joined, logic tutorials completed).

Challenges we ran into

Voice & Symbolic Logic Integration: Balancing free-flowing AI conversation with strict logical validation required multiple checks and fail-safes.
Real-time Performance: Ensuring audio reflective particles and voice outputs remained smooth under multiplayer conditions.
Tutorial Design: Making symbolic logic approachable and enjoyable for beginners, yet still useful for advanced debaters.
Scalable Infrastructure: Handling voice processing and logic checking with minimal latency on the backend.

Accomplishments that we're proud of

Seamless Voice + Logic: Successfully merged AI voice responses with an internal symbolic logic engine—a first of its kind for us.
Interactive Tutorial: Built a user-friendly introduction to five fundamental logic symbols, helping users quickly improve their reasoning skills.
Engaging Social Platform: Launched a multiplayer feature that encourages friendly debate, skill-building, and community interaction.
Achievement Gamification: Integrated a rewards system that keeps users motivated and tracks their logic proficiency over time.

What we learned

The importance of careful UX design for bridging voice interaction and formal logic, ensuring neither overwhelms new users.
How to make LLMs and a formal logic engine collaborate effectively—LLMs handle natural language, while the symbolic logic system ensures semantic correctness.
Techniques for optimizing real-time audio visualization without sacrificing system performance.

What's next for VeriVox

Expanded Logic Library: Add more advanced logic operators (e.g., quantifiers, modal logic) for deeper debates.
Community & Collaboration: Let users share logic maps, team up for group debates, and crowdsource best practices.
Multi-Language Voice Support: Adapt the platform for multiple languages and diverse cultural contexts.
Advanced AI Opponents: Develop more sophisticated voice-based AI adversaries with specialized knowledge in specific debate topics.

Built With

elevenlabs
figma
flask
openai
python
react.js
restfulapi
touchdesigner
typescript
websockets

Submitted to

ElevenLabs x 16z Worldwide Hackathon

Created by

I worked on UIUX Design, co-work with front ended engineer, and set up audio active particle visual. I also work with demo of interaction of VeriVox. Editing the Project video and Product Management. Designing the whole user experience and user research + market research.

Xiaoyang Wang
Georgia Tech Pscy&ID. UIUX, Product, Research, Cognitive Science
I led the front-end development of our AI-driven platform, focusing on creating a seamless user experience and enabling real-time interactions. Using TypeScript and React.js, I built core components for the Lobby, Tutorial, and AI Agent interaction interface. I also integrated RESTful APIs so users could easily click to record and send voice inputs to the backend models, enabling multi-turn AI conversations powered by services like ElevenLabs TTS. The project expanded my understanding of AI’s broad applications and showcased how AI delivers a user-centric experience. It was thrilling to merge voice processing with large language models, turning our own AI ideas into reality!

Cecilia Qian
I architected and implemented the FastAPI backend system, including RESTful APIs, WebSocket services, and data models that seamlessly integrate voice processing (ElevenLabs TTS and Google Cloud STT) with multi-turn AI agent conversations. Also I led the backend development by collaborating with team members responsible for frontend and NLP to create a robust infrastructure for real-time debate interactions and user session management. This is the first time building AI Agent, so cool!!!

Liu Xuanzhang
I designed and implemented the NLP component for the AI Debate Agent project, delivering robust logic analysis functionality. This component converts user input into structured logical expressions and performance metrics in JSON format, facilitating seamless integration with both the speech-to-text module and the front-end visualization system, and thereby laying a solid foundation for the project's argument analysis capabilities.

Sydney Chen