Show HN: I built an AI book recommender in 2 days

Hi HN! I built this in ~2 days over the New Year's break, and it was the most fun I've had coding in a while...

*Why I built it:* I was tired of generic listicles and recoms from Google Search, ChatGPT/Gemini, and Goodreads. I wanted something where I could say "a cozy mystery for a rainy weekend" or "something like Kafka but less depressing" and get actual niche results.

*How it works:* - Describe what you're looking for in plain text (or voice) [https://mynextbook.ai/onboarding-books] - A RAG system, built using Gemini 2.5 Flash Lite + Exa search, finds personalized recommendations - Metadata is fetched from multiple sources - ~5-10 seconds from input to recs (5-10X faster with cache)

*Stack:* - Next.js 14 / TypeScript on Vercel - Neon (Postgres) + Prisma - Gemini + Exa Search API - Built with Cursor + Claude Code, & AntiGravity (happy to discuss which was better for what)

*Try it without signing up:* no sign-up required to explore

*Bonus:* It also recommends movies and TV shows based on your book taste! I've found some binge-worthy stuff through it that I would've never found otherwise.

Still a tiny side project, but I love seeing people actually find useful recoms with it. Would love feedback on rec quality, UX, or features you'd want!