 Hello, everybody. My name is Jorge Torres. And MyCB brings the entire ML ecosystem to databases. And let me walk you into that. Also, the problem that we're trying to solve is precisely what they were describing before. It's the idea of being data-centric as opposed to machinery model-centric, which is probably the way that most people do it today. And that has a lot of problems. The first problem that you have is that it carries a lot of the intuitions that we bring in from research. But when we want to apply it, it's not necessarily the case. So when you're doing modeling and you forget where the data is coming from, you start dumping a lot of files into your computer and make a lot of transformations in Python, et cetera, how you want to transform. And eventually, you have a model that works. But it has a lot of issues. The first issue is that now you're going to have to turn all of that spaghetti code that you built to transform data for your model into some ETL-ing tool. And now you have to maintain both. And if you deploy your model, usually people deploy these models behind like a RESTful API or some other web service. And every time you want to make a prediction, you need to extract data from the database, run the ETL-ing pipeline, run the model, and many times have to put the predictions back into a database of the data warehouse. So there's a lot of unnecessary ETL-ing happening. And what Miceby tries to do is to simplify the problem. Essentially, if your data is in the database, you can train the model directly from the database. And actually, you can consume the model as if it was a table. I think that is the second innovation that Miceby brought to the table. And why Miceby today is the fastest growing applied ML project in the world. And the concept is very simple. Imagine that you have a table, this simple table. Of course, tables in production don't look like this. But for the example, you have income and debt. And databases are meant to give you a response if what you query has a direct row or an exact match into a database. So say, for instance, if you want to ask, what will be the debt for when income is $80,000? And if that exists on the table, then you will get a response. But they're not designed to give you an approximation. Now, anyone that has done machine learning knows that you can build a hyperplane that can fit that data, a machinery model. And what Miceby allows you to do is to create that model in a syntax that is very similar to that of creating a table or a view. So you can say, create model, the name that you want to give the model. And then from what table you want to learn and what column you want to learn. Now, you can specify what machine learning network or what machine learning capability you want to bring into the mix. But if you don't specify anything, then Miceby will figure this out. Now, the cool thing is that now you can query that model as if it was a table. So you can say, OK, tell me what will be the income and debt and the predicted debt for when the debt model is, I don't know, $90,120. And then you get that approximation. Now, again, most problems in machine learning are not as simple as this. The concept is a concept that extrapolates all the way to very sophisticated stuff. So one example of this is NLP. You have a growing ecosystem of machine learning frameworks like Huggingface. And with Miceby, you can bring the power of Huggingface down to the data layer. You can query models that are meant to be doing NLP and outside of the database in the database itself. For example, imagine that you have thousands of reviews of products, and you want to get sentiment about what people think of your products. Now, Miceby does this in a very simple way. You can say, I want to create a model, however you want to name the model, and you want to predict sentiment, and then you use the engine Huggingface. And the reason why you have to specify the engine is that now Miceby supports more than 70 engines of machine learning and more than 70 data frameworks. And then you specify, OK, what will be the actual model that you bring from Huggingface, in this case is a Roberta classifier. And then what is the column that you want to predict from, and what are the tags or labels that you want to do? Now, one single SQL statement brings the power of Huggingface to any database. And again, that any database by that we mean Miceql, Postgres, Oracle. Now, once you have this model as a table, it becomes very easy for you to query that model like if it was a table. So you say, OK, let me ask the sentiment classifier, what will be the sentiment when I have a sentence like this one? And of course, for the sake of this example, that sentence is very stupid, but you get the concept here. It's like neutral. Now, the second thing is you can play it the same way, like what will be if the sentence changes to something positive. But the interesting thing becomes when you actually want to do bulk operations. Say, for instance, you want to get the prediction for the sentiment for a whole bunch of rows. Again, imagine all the products that you have. Let's say that you have this table. You can join the table with the predictor, and then you get a prediction of the sentiment for each one of those. Now, again, the concept that I want to bring to you is that Miceql is bringing this capability to all of the developers. At the end, only 1% of the developer population knows how to really implement effectively research type of machine learning. And what we do is we bring the state of the art of machine learning to the people that actually know how to build products. And well, I invite you guys to go check it out. Miceql.com. Thank you.