I’ve always felt that most book recommendation engines are shallow, they either use tags or “people also liked” data. I wanted something that actually understands what you mean when you say “slow burn sci-fi with emotional depth” or “books like Harry Potter.”
So I built FindReads . It uses GPT to interpret your prompt, queries for books, and returns real titles that fit the vibe.
Stack: Next.js, OpenAI API, MongoDB, deployed on Vercel.
It’s still a work in progress — the results are surprisingly good for some moods, weird for others, but I’d love feedback from the HN crowd on how you’d approach improving it.
Things I’m exploring next: • Combining embeddings + genre metadata for more consistent results • Ranking by user upvotes • Fine-tuning prompts for better diversity in suggestions
You can try it here: https://www.findreads.app
Would love to hear how you’d make something like this more reliable or fast.
Show HN: FindReads – I built a site that recommends books by vibe
findreads.app