What makes this interesting: 100% AI-coded project. I didn't write a single line of TypeScript – just acted as "AI whisperer" guiding Claude 3.7 and Claude 4 through architecture decisions and debugging.
This project isn't supposed to be useful – it's an experiment to test AI agents to their limits and see what emerges from pure human-AI collaboration.
Key highlights:
182/182 tests passing across monorepo packages Dual build system (ESM + CJS) Real-time AI character conversations with TTS Glass morphism design that exceeded expectations Multi-language support (6 languages) The journey showed how AI coding evolved: Claude 3.7 needed constant supervision (creative but prone to "hallucinations"), while Claude 4 became a trusted partner delivering production-ready features autonomously.
Tech stack: Next.js 15, TypeScript monorepo with Turbo, LangChain, OpenAI integration, GitHub OAuth, Vercel deployment.