Disclaimer: Large parts of the codebase have been written with AI. Read more on that in Vibecoding.
Bookhunt is a website that makes public bookshelves (sometimes calle free little libraries) searchable. You can search for books or authors and see in which bookshelves they have last been spotted. You can also simply browse recently seen books in bookshelves nearby.
It is just as easy to contribute. Simply snap a photo of a bookshelf, upload it and Bookhunt will extract the visible books and add them to the database.
All data is located in public/data.
Proprietary books and bookshelves data is being made available under the CC BY-NC-SA 4.0 license.
Additionally, third-party data gets sourced for bookshelves:
public/data/bookshelves/bookshelves_bal.json: Sourced from www.boite-a-lire.com. Provided by github.com/Binnette/bookcases-boite-a-lire under CC BY-NC-SA 2.0 license.public/data/bookshelves/bookshelves_box.json: Sourced from BOKX / urbanlife. Provided as Google My Maps.public/data/bookshelves/bookshelves_clc.json: Sourced from croquelivres.ca.public/data/bookshelves/bookshelves_obc.json: Sourced from OpenBookCase. Provided under ODbL v1.0 license.public/data/bookshelves/bookshelves_osm.json: Sourced from OpenStreetMap using overpass turbo (Europe only). Provided under ODbL v1.0 license.public/data/bookshelves/bookshelves_wee.json: Sourced from Westenergie. Provided as Google My Maps.
This project serves me as a sandbox to play around with vibecoding. I largely don't use integrated code generation assisstants in my professional work, so this is a project where I can freely experiment without needing to worry about code quality or security concerns.
Almost 100% of the codebase has been generated by LLMs.
I chose this application because it's an easy enough use case and comes with a lot of boilerplate frontend code which I don't enjoy writing myself at all.
Some learnings (nothing surprising here, most of you probably already know this):
- LLMs are really good to quickly create working prototypes so you can prove a market fit early on.
- Quality seems to be good enough for what it is.
- With a large enough codebase, the LLM will stumble over itself and will for some issues never be able to fix them, even though it already proudly declared to have fixed them 10 times.
- I am concerned about continuing to maintain this on my own. I do have a pretty good understanding of the system architecture since I dictated it, but I don't know the code and would probably struggle to change stuff on my own now.
- I dislike that you have no direct visibility of costs. You vibecode away and lose every feeling for how many resources you currently consume (regarding both financial and environmental costs).
- Availability of LLM APIs is pretty bad.