Open Source Neural Search Framework Jina AI - MLOps Discussion ft Alex C-G

Michael
Okay, I think we're going to start it up. Let me check to make sure it's actually working... looks like we are indeed live—great. I'm going to try to check my volumes. I know that every time I do one of these, my voice just explodes through whatever mic I’m using, so I’ll try to keep it a little Bob Ross this time—calm. Which is totally not me, but whatever.

Thanks everyone for joining. We are lucky enough to have the Developer Relations Lead over at Jina AI, an open source neural search framework: Alex CG. Is that how you introduce yourself, by the way?

Alex
Yeah, my parents are terrible people. They gave me a really long double-barrel surname. I sound really British, but I’m not—I just go with CG.

Michael
Nice. What’s the full one?

Alex
Griffiths-O’... yeah.

Michael
Wow. French and Welsh?

Alex
Why not—two great tastes that taste terrible together.

Michael
No comment on that one. Welcome. A couple things to run through and then I’ll kick it to you. I think our audience today will mostly be from the Label Studio community, but there may be folks from Jina AI as well.

For anyone who’s not already in our Slack channel, please join—especially if you want to ask questions. Our engineering team is extremely responsive. We're all about community here, so it's worth jumping in.

Also, we have an updated webinars page on our website. You can RSVP for any upcoming sessions and get calendar reminders. We're adding more events and talking with other exciting ML and MLOps communities, so expect that list to grow. You can also watch replays of past webinars on YouTube. Note: livestream links expire—we trim the videos and repost them, so check the webinars page for working links.

Obviously, I’ll let you introduce your tool, but if people want to follow along or start queuing up questions as Alex talks through Jina, you can visit get.jina.ai to check out the GitHub repo.

And if anyone here isn’t familiar with Label Studio, I’ll do a quick overview—something I haven’t really done much of in previous webinars, but I should. Label Studio is an open source data labeling tool. You can label pretty much any data type—with more supported all the time. You can check it out at labelstudio.io, try out the Playground to test different setups, or explore our open source and commercial offerings.

We have an Enterprise Edition, a Teams Edition, and a free Substack newsletter at labelstudio.substack.com. Of course, you can always visit our GitHub repo from any of those links.

Okay, that’s all of my plugs. Over to you, Alex!

Alex
Thanks, Michael. As mentioned, I’m the Developer Relations Lead at Jina AI. We’re a startup based in Berlin—because what other Berlins are there? Probably not many with AI startups. Anyway, we provide an AI-powered search framework that’s fully open source.

I’ll walk you through how it works, what it can do, and show some code if we have time. It’s surprisingly simple, even if you don’t have a background in deep learning. I majored in Chinese at university—everything I know about Python and AI I learned on YouTube. I’m no expert, and thankfully, I don’t have to be. This stuff is powerful but pretty easy to use.

Let’s say we want to build a search engine for memes. You know the Winnie the Pooh meme—the one where he's wearing a red shirt in one panel and a tuxedo in the next? Let’s say we want to find memes about Winnie the Pooh and food. In the example meme search engine we built, that’s exactly what we do.

You get memes with the words “food” in them—like “finger food” or “hors d'oeuvres”—but also memes that mention cereal or breakfast soup or cheese puffs. The model understands those are food-related, even without the word “food” explicitly there. It’s using a language model pre-trained on a bunch of data to make those associations.

You can also search by one keyword like “chocolate” and get a wide variety of results. You can try this at examples.jina.ai—although fair warning, I think the back-end fell over again, so I might need to kick the server later.

But that’s the idea: a working search engine where you can input text or images and find similar memes. That’s what Jina is built for—searching across all kinds of data.

Michael
We’ll stop here. Want me to keep going with the rest of the transcript in this format?