The Reality Check: I’m not a professional dev. I’ve done my best to get this to a working prototype, but my best is likely your worst. I’m sharing it here because I’ve hit walls I don’t know how to climb yet.
The Hard Problems:
The API Key Barrier: Requiring an OpenAI/Gemini key is too much friction for normal users. How do we move toward local LLMs (Transformers.js?) or a community proxy?
The Trust Engine: How do we weight sources without building in our own bias?
Performance: Parsing real-time transcripts and grounding them in search results without lagging the browser.
I've set up the repo with an MIT License, CI/CD, and a roadmap of "Good First Issues." If you've ever wanted to use your skills for factual integrity, please take a look at the code and tell me where I'm wrong.