Machine learning is fast becoming a ubiquitous technology—but if you don’t fully understand what it is and how it works, you’re not alone. It’s an evolving idea with complex functionality.
But our guests on this edition of UpTech Report figured out a way to conceptualize machine learning in a way that is easier to understand and apply in real-world settings.
Jorge Torres and Adam Carrigan are the co-founders of MindDB (CEO and COO, respectively), a company that treats machine learning as a database layer, allowing you to retrieve predictive analysis by performing a simple SQL query. With their technology, obtaining insights on future sales is as easy as checking your inventory.
On this edition of UpTech Report, Jorge and Adam explain how their implementation of machine learning works and its intriguing potential.
More information: https://mindsdb.com/
TRANSCRIPT
Jorge Torres 0:00
If you start thinking of machine learning as another representation of your data again as a as a as a table, then you can seamlessly integrate predictions and data that is historical.
Alexander Ferguson 0:20
Welcome, everyone to UpTech Report,our Applied Tech series. Today’s episode is sponsored by TeraLeap. Learn how to leverage the power of video at teraleap.io. My guest today are Jorge Torres, and Adam Carrigan, San Francisco and UK, co founders of MindsDB. Welcome, guys. Good to have you on.
Jorge Torres 0:39
Great to be here.
Alexander Ferguson 0:40
Now their product is an open source AI layer for databases. If for those other if you’re running queries on databases, maybe a developer data analyst, this might be an intriguing tool and platform to check out on your site. It says machine learning straight in databases through AI tables. I’m curious, how did you guys come across this problem initially and say, We got to solve this, let’s build a solution.
Jorge Torres 1:06
We we saw that there’s two branches of machine learning, there’s kind of like research in machine learning. And then there is applied machine learning. And for the latter people that are more interested in applying machine learning techniques, really the main ingredients data and and you just take algorithms that you know work for different data types, and you apply. And given that the main ingredient is the information that you fit into this algorithms. And a lot of this data already exists in databases, then it makes sense to treat machine learning as an as an actual data layer component. To be more precise. If you start thinking of machine learning as another representation of your data, again, as a as a as a table, then you can seamlessly integrate predictions and data that is historical, in with your applications, or whatever your use case is. And in that line of thought we decided that the best way to summarize it for someone that is used to working in a database is to express the mechanics of machine learning through a concept that we call AI tables. And the main difference between an AI table and a regular table is that when you query a regular table, you’re essentially just looking for precise values that you have in your table. When you’re querying a table, you can get predictions in return. So essentially, a generates data up and being queried. And his data, of course is or the way that it generates this data is based on the information that you have in your database itself.
Alexander Ferguson 2:52
This concept that it moved to, we need to solve it what what was the first kind of couple use cases, if you give me a someone that you saw using? Now this solution? How does it change anything that they’re normally
Jorge Torres 3:07
doing? Yeah, so usually, when you’re building an application, in machine learning, you have your machine learning application that has to go through different stages, you first develop the model, you train the model, to train the model, you have to extract data. And then once you have the model trained, when this model has been consumed from like, you know, your solution, it has to go again still to the database, pull some data, send it to the model, the model then gives you some return kind of predictions, you aggregate this information go back into the application. So essentially, you have like this triangle of information going through like different stages of your application. In a traditional manner, you only have your application and your database, and information goes back and forth. So for a developer that uses might be this remains the same. To give you one precise example say for instance, you have a website that deals with, you know, on online retail, and you want to predict demand for a given product. Ideally, what you would like to do is you want to select from the tables that have your products and your current inventory. And you will like to predict what will be the inventory for like the next week or so. And if you could do this straight from a query, then for the person that is building these applications, it will be the same way that you’re querying your information. Nonetheless, now you’re getting predictions and results. So that that is one of the the advantages of what we do essentially the same language, same process that you go at building any normal application. It’s just that now you have predictive information and in combination with historical information.
Alexander Ferguson 4:54
So like being able to do this existed, but it was a little more complicated and a two step or multi step process. As in this is simplifying and effectively and keeping it all in the same place that is that I get that correct?
Jorge Torres 5:05
Yeah, yeah, it’s all about simplification. And the main intuition behind it is that when you’re applying machine learning, what you really want to do is to shrink the development lifecycle of traditional research, machine learning. And right now, many departments have kind of carved out data science teams that work independently from the machine learning from the engineering teams. Mostly because believe me likes development life cycles are so much lower. But developers and and people that build tools, through through data, they are used to just, you know, the Agile form of development. So what we do essentially allows you to fit any machine learning task that involves structured data into the regular software development, development life cycles that companies may have.
Alexander Ferguson 5:57
So Adam, what’s on the roadmap? What are you planning for my CV that you’d want people to know about?
Unknown Speaker 6:03
Yeah, so one of the big things that we’re working on over the next sort of several months is tighter integration with with the databases. So we’re working with a number of organizations, both for profit and not for profit, to have tighter integration with with these databases. And so that means, you know, as Jorge mentioned, that it becomes a much more streamlined process, and that these communities, one example of one, one database that we can mention is Maria dB, one of our investors is actually the former founder of Maria dB, and so working, working very closely with with their community to, to bring this integration to their to their database.
Alexander Ferguson 6:41
What’s the business model that you guys have in place and says, There’s no place to buy anything? So what’s the future business model for this?
Unknown Speaker 6:49
Yeah, so if you’re a larger organization, there are certain things that that you need to be able to run mines dB, and scale it, if you’re a smaller organization, the open source edition is completely fine for you. But once you need to scale up, you just start your clustering, you need authentication security, then we have an Enterprise Edition. And that obviously comes with support and consulting and sort of the usual services that will help these organizations actually run it at scale.
Alexander Ferguson 7:17
What’s a word of wisdom for a developer or data analyst, when they’re working in this type of environment that you’d want them to share? Because you’ve probably been hearing this questions, you’ve been working on this problem, but a word of wisdom that you could share.
Unknown Speaker 7:34
Yeah, really, that anybody can do this. In a previously machine learning and AI sort of model creation was really just a domain of very highly skilled data scientists. And the message that we’re trying to get across is that really, anybody can do this, whether you’re a developer, you’re learning to code, whether you’re a data analyst or database admin, you can do this yourself. And you can build these models, you can move them production, very, very easily with with nice dB.
Alexander Ferguson 8:01
For those who want to learn a bit more about the story. Join us for part two of our interview where we’re gonna hear a bit more of the journey that they’ve been on, but to give a taste. This has been about three years in and asked both of you, what would you wish you knew three years ago that you know now, or I want to share first, what what would you wish you had done three years ago that you know, now?
Jorge Torres 8:22
I think that we knew nothing about open source in general. So we’ve we’ve been learning everything the hard way. But to summarize into one thing, it would be that there are many ideas that you have, that will require a lot of user validation. And the best way for anyone to get those ideas out there, we’ve learned is to go through the open source route, mostly because the open source community since they’re not expecting to pay at the beginning, they’re the best way to get feedback into the solution that you have. And we were always going to be grateful to everyone that has contributed in the open source side of mind to be purely because it went from an idea that Adam and I had all the way into something that, you know, 1000s of people use, that could not have been possible without having engaged in the open source. The second thing probably has to do with just because a business is open source doesn’t mean that can’t be a business, but we’ve learned is that you can have a great user base and the open source community and still find a way to make it sustainable, and make VC bankable and whatnot. So those two probably are the biggest insights.
Alexander Ferguson 9:47
Adam for you.
Unknown Speaker 9:48
Yeah, I think one of the biggest things that we’ve learned and I wish I knew a few years ago was that almost certainly there has been someone or some some company that has faced a problem before You know, when we were sort of very early stage, you know, we were, we were talking internally about how to solve many problems that come up with, with running a startup. And it wasn’t really until we kind of started to engage more with the startup community in San Francisco, we participate in Y Combinator and Skydeck. And really talk to a number of grit Advisors, a number of startup founders that have been through this journey before. And they could give great advice, you know, nerve situation is identical, but they have probably been through something similar. And reaching out to these people and getting their opinion is something that you know, we now do on a regular basis, both advisors investors, you know, just people we met for coffee, but I wish we would have done that a little bit sooner.
Alexander Ferguson 10:49
I appreciate both being able to break down. What is MindsDB where you guys playing a role as well, some insight, definitely check out part two of our interview. bit more about their story. Thanks again for joining us. That concludes the audio version of this episode. To see the original and more visit our UpTech Report YouTube channel. If you know a tech company, we should interview you can nominate them at UpTech report.com. Or if you just prefer to listen, make sure you subscribe to this series on Apple podcasts, Spotify or your favorite podcasting app.
