-
An Example Query
-
The Home Screen
Inspiration
The search interface of libraries is often confusing and overloaded. Also, it is usually only possible to search for specific things like a title or an author, not for a general description of a book. Addressing this was exactly our vision: Building a unified, simple interface that allows to search books available at Stadtbibliothek in every way imaginable.
What it does
BibXplore allows you to browse the Stadtbibliothek München's books blazingly fast by asking questions. A book on dragons and castles that has less than 200 pages? All Fantasy books about Hamburgers? The only limit is your imagination. BibXplore can answer all of these questions.
How we built it
For the frontend, we used react and tailwind. For the backend, we used Gemini, Gemini's embeddings, SQLite, and the vector index HNSW. All books were embedded so we can search them semantically. Additionally, the lightweight database SQLite provides support for specific and numeric queries, such as filters on the number of pages.
Challenges we ran into
The information provided by Stadtbibliothek is incomplete: For many books, they do not have information such as the genre. We queried other APIs like google books too - however, they often don't contain the German books in Stadtbibliothek.
Accomplishments that we're proud of
In this limited amount of time, we came up with the idea and were able to implement a working prototype.
What we learned
This was the first Hackathon for both of us. Hence, we learned a ton about ideation, rapid development, and pitching.
What's next for BibXplore
BibXplore is a Proof of Concept - it does not contain all books of the Stadtbibliothek due to the aforementioned incomplete information. With more time for preprocessing, one can use a larger number of webcrawlers to obtain missing infos from sites like Amazon, and add the missing books. Then, the site could be hosted to help all inhabitants of Munich who like to read.
Built With
- gemini
- hnsw
- text2sql
