I built a proof-of-concept platform where multiple AI personas (powered by Claude, and eventually other LLMs) discuss technical problems from different perspectives. Think of it as a virtual engineering council where each AI takes on the personality of both a famous engineer (like Linus Torvalds or Grace Hopper) and a specific technical mindset (pragmatist, theorist, etc.).
The idea we are exploring here is that if we challenge the AI to take on different personas, it will prevent the AI from getting entrenched in its own opinion.
How it works:
- Put in your Anthropic API key (the whole app is frontend only, so keys dont travel anywhere other than your browser's local storage)
- Choose your council's profession (Engineering, Design, Product, etc.)
- Input your technical problem or question
- Watch as 3 AI personas discuss and debate the solution
- Each AI takes turns responding, building on previous answers
- Configurable number of turns per participant
Please note: The names of the personalities are fictional and don't have any effect on the discussion. Just there for some fun.
It is open-source so any and all changes, requests and feedback are welcome. But it will be highly preferred that the PRs come from code written using Cursor or Github Copilot or matching that code quality in terms of readability and maintainability.
Demo: https://the-council-of-perspectives.vercel.app/
GitHub: https://github.com/yashodhanmohan/the-council-of-perspective...
This is a very early prototype, but I'm curious to hear the community's thoughts on the concept of multi-agent AI discussions for technical problem-solving. Would love feedback on both the implementation and the broader idea.
