RhetoricAI

Speak up for yourself!

Comment

Inspiration

In a world overflowing with information, manipulation and fake news are commonplace. That’s why, beyond resisting external influence, it’s equally important to communicate with clarity and precision. We believe that refining rhetoric is crucial to avoiding misleading interlocutors and committing logical fallacies.

Problem

Lack of accessible tools for learning and improving structured and persuasive debating skills.

Solution

A web application powered by AI that can assess users' current skills, suggest improvements based on a given speech, and hold conversations with them.

What it does

RhetoricAI allows users to engage in conversations with the Debate Coach. Thanks to AI models, it can analyze users' weak points, evaluate their speech, and provide advice on improving their speaking skills.

Tech stack

  • React UI created with Lovable
  • FastAPI (Backend)
  • Whisper by OpenAI (Speech-to-text conversion)
  • Conversational AI by Eleven Labs (Debate Coach)
  • Mistral AI (LLM engine)
  • Deployed with Koyeb

Challenges we ran into

  • Integrating multiple APIs and external software
  • Ensuring smooth UI communication with the server
  • Response sanitization
  • Refining the final concept

Accomplishments that we're proud of

  • Deploying the MVP
  • Designing the UI
  • Coming up with the main idea

What we learned

  • Collaborating on a complex project as a team
  • Using new AI tools for tasks we have never done before

Future Concepts

Personal assistant

A wearable device that analyzes daily conversations and provides feedback and fact-checking at the end of the day.

Training mode

Training software offering live practice with virtual opponents featuring customizable parameters and discussion styles. Provides feedback with suggested improvements.

Real-time 1v1 debate

An AI-powered system that analyzes debates in real time, evaluates arguments, assigns points, and highlights logical flaws. Particularly useful for political debates and media interviews.

Fine-tuning

Enhancing the LLM with user-generated input and publicly available datasets, such as Oxford-2B and Cambridge-2B, to improve its understanding of debates.

Built With

Share this project:

Updates