[Koha] Integrating Vision AI / GPT for cataloging in Koha

Tue Sep 24 04:52:52 NZST 2024

Hello Arthur,

And to everyone else who wish to try,

You can now test this with me here 😁:

https://marcai.libmatic.com/

<https://marcai.libmatic.com/>
For Unimarc, I think it is not a problem ,we can modify prompts
accordingly, but I didn't try that yet, I added this as a todo item so I
can try it later.

For now to use the tool you'll need to:

   - create an account,
   - verify via email
   - configure organisation details ( including OpenAI key )
   - (optional) configure Koha link, api key and secret, to send bibs to
   Koha once happy with it.

This is multi tenancy setup, so whatever you submit will only be visible by
your organisation only,

For Gemini, the current published version is not utilizing it at the
moment. I have code ready for that, but it requires some enhancements,
which I will work on later if others are interested in using it as well. I
find Gemini to be effective for non-Latin texts, which is why I chose it,
but generally, OpenAI is excellent for English and Latin-based languages.
so even for later Gemini is optional.

For vector stores, this idea is related to another concept that can build
upon our current work with the library catalog. To improve cataloging, it’s
essential to implement a Retrieval-Augmented Generation (RAG) system to
retrieve all relevant records in the library catalog, ensuring consistency
in handling similar materials. Using embeddings and vector-based stores is
ideal for this purpose.

Another idea driven by having the catalog indexed as embeddings and having
RAG ready,  is to create a library chatbot that is aware of the library’s
content and can communicate with the Koha API. Users will be able to ask
the chatbot to find books on specific topics (patrons can ask in any
language), and the AI will return related items. Since the AI understands
the library’s collection with access to Koha Rest API, it can inform users
whether books are available or checked out. Users may also request the AI
to reserve a book, and it can handle the reservation for them or present a
button for confirmation. The same functionality can apply to renewals and
any other actions that users typically perform manually and that is
supported by Koha RestAPI.

I believe there are many ideas to explore and experiment with in this
space. So, let me know, everyone, if you have any questions, comments, or
suggestions for improvement. I’m all ears!

Karam.

<https://marcai.libmatic.com/>

On Mon, Sep 23, 2024 at 8:20 PM Arthur Suzuki <arthur.suzuki at biblibre.com>
wrote:

> Hello Karam,
> I would be interested to get more details from your project, or make
> some tests of your tool.
> Since you don't share the sources *yet*, is there anyway to try out your
> tool? maybe available as a website or as a pre-compiled binary?
> I've also seen you use OpenAI or Gemini, but you also mentionned some
> vector store, does it mean a company would need to host a vector store
> to make it working?
> Or can it work with only OpenAI paid account?
> What about compatibility with other MARC formats? I'm thinking about
> Unimarc.
> Best,
> Arthur
>
> On 2024-09-19 19:35, Karam Qubsi wrote:
> > Hi Mike,
> >
> > can you share more details about your project?
> >
> > At the moment the project is a tool required for a research paper I was
> > invited to participate in with librarians in the University of Nizwa
> > library ( Oman),
> >
> > This tool is a must for this research, to have better integration
> > between
> > AI and the Library system there (koha) and gauge cataloging stats
> > before
> > and after and evaluate AI benefits to cataloging in both quantity and
> > quality.
> >
> > The project is not sponsored by the library or the university , it is
> > just
> > volunteer work from me so far, to build this tool, I found it
> > interesting
> > and enjoyable.
> >
> > Is it open source project?
> >
> >
> > I love opensource, but unfortunately, it is not open source, or *at
> > least
> > not yet*,
> >
> > but anyway at this phase as it is still in development I'm not going to
> > charge anything, so totally free to try it, if any library here would
> > like
> > to try this using their own openai key / or gemini key,( you can get
> > free
> > credits from both ) then I will be happy to help to integrate this tool
> > with their library. ( get in touch if interested ).
> >
> > Finally* I may make it open source* if proper support is provided, like
> > if
> > some company would like to support making this opensource by hiring me
> > to
> > work on it as an open source project 😉, or if a library or few
> > libraries
> > can sponsor the project, I will be more than happy to make it open
> > source
> > then,
> >
> > If not then the project will need a way to support itself and to
> > sustain
> > and grow,
> >
> >  Is it possible to test it in my natural language?
> >
> >
> > Yes, as mentioned if you are interested, I'll get in touch with you
> > within
> > a few days, so you can try it with your own Koha system. I will host it
> > somewhere first so you and other libraries can try it.
> >
> > Thank you.
> >
> >
> > On Thu, Sep 19, 2024 at 11:06 PM Mike D. <black23 at gmail.com> wrote:
> >
> >> Hello,
> >> can you share more details about your project? Is it open source
> >> project?
> >> Is it possible to test it in my natural language?
> >>
> >> Thank you. Regards
> >>
> >> Michal
> >>
> >> čt 19. 9. 2024 v 15:15 odesílatel Karam Qubsi <karamqubsi at gmail.com>
> >> napsal:
> >>
> >>> Hello Koha community,
> >>>
> >>> I’ve been working on a side project experimenting with AI, and I’ve
> >>> built
> >>> something that looks interesting to me. I’d love to see if other
> >>> members
> >>> of
> >>> the Koha community are interested in something similar.
> >>>
> >>> You can watch this 35-second demo I recorded:
> >>>
> >>> https://youtu.be/vnvxmjTUmGQ?si=-YHqLoyp0VeqRL6L
> >>>
> >>> If you have any questions, feel free to ask.
> >>>
> >>> Thanks!
> >>>
> >>> --
> >>> *Karam Qubsi*
> >>> _______________________________________________
> >>>
> >>> Koha mailing list  http://koha-community.org
> >>> Koha at lists.katipo.co.nz
> >>> Unsubscribe: https://lists.katipo.co.nz/mailman/listinfo/koha
> >>>
> >>
>
> --
> Arthur Suzuki, 🌈🏔️
> Développeur @BibLibre
>

-- 
*Karam Qubsi*