10.25.23-Digital-Einblick

PAUL YANG: Hi, everyone. Sorry to stand between you and lunch, so we'll keep this hopefully nice and tight. Einblick is a multiplayer data canvas, and one of the things that we have done, actually, is to make generative AI a core component of our platform. I actually think Dr. Sanchez's talk before this really motivates this conversation a little bit.

Green button. Oh, this one. And I think the problem is that data science today is not necessarily collaborative, and it's definitely not effortless. It looks a little bit like this, where people look at a big screen of code, and they're a little bit unhappy.

If you actually go on Google, this is true. One in five Google search results is about failing, right? And this is a fundamentally broken way to approach a problem. 98 things that can go wrong with your ML project is too many things.

And one of the core problems is this kind of data democratization question that has bounced around for decades. And today, a really good data team might look something like this. A data engineer emails the data scientist, The data is ready, good luck. The domain expert might come in, have a couple of chats, and also say, Here's some data, good luck, some context, good luck. And then the data scientist might use Jupyter Lab or another data science notebook to produce their solutions, they use a PowerPoint to share it with the decision makers, and you go through this iteration where, oh, you forgot something, let's take it back to Jupyter.

There is not a shared environment in which everyone can communicate. And one of the exciting things is that Einblick puts the user experience first by putting it onto an interactive canvas. Miro is successful at being a project management collaboration platform. Figma has taken the design world by storm.

But I think that data science is not inherently less collaborative than any of these other platforms. And so that's why it was necessary to take the traditional Jupyter Notebook and put it into this non-linear canvas that allows you to manage execution, bring in your whole team, and lay out your analysis visually.

But beyond that, one of the things that we are able to do-- this was a short demo-- is integrate a large language model as a primary way to interact with your data and create outputs. One of the things that I think I'd like to point out is that it used to be that translations were really bad, and you could not rely on Google Translate to do your Spanish homework.

But now you can ask questions like, I'd like to order a pumpkin spice latte, and you can translate it into pretty much any language you want. But there's not really a reason why it could do this sentence into Welsh, but not your desire for a particular chart that looks a certain way into Python.

And so one thing that we're able to do is-- oops-- one thing we're able to do, and I encourage you to come by the booth and try it out or watch us try it out, is kind of just say, Hey, I'd like a correlation matrix of my data, and get this result back. There's no need to do the tedious coding in syntax.

People have a much better idea of what they actually want than the ability to necessarily write code all the time. And so this is going to be a massive unlock for organizations which have small amounts of data, medium amounts of data, and more people who would like to access this data, but don't necessarily have the language to do so.

As an example, we have done a user study. And essentially, one of the things that we see is that even for expert data scientists, we can strictly reduce the amount of time you take across the tasks that are common, including visualization and model building.

And part of this is just coding is about speaking with perfect grammar, and unless you all got 800 on the SAT writing section, we all know that that's not a natural way to communicate, right? And so the natural language model and the natural language interface allows our users to produce the outputs without necessarily having to go through the process of writing perfect syntax.

As a case study, we were deployed to a chemical company. And essentially, what we offered them was two main improvements. The first one was the ability to grab the whole team and pull them into the canvas at the same time. We actually joined-- we were a part of this experiment, and so we also helped accelerate that. But essentially, the data scientists brought their lab-- not lab, it was a factory manager in.

And when you think about a factory, right, the person who's managing the factory is a human learning model. They've stood in front of this panel of dials every day, and they look at this set of conditions, and they say, Hey, if I see these dials look like this, the temperature is like that, I know that I'm going to have a shutdown event. So this human was able to manage that factory.

But that's a lot of human knowledge that is not necessarily embodied in the data itself, in the IoT sensor data. And so by bringing those stakeholders together into a single canvas and being able to build on the fly, they were actually getting to initial ML model within a week.

And one of the arguments we make is, hey, it may not be the perfect model, but if today you're not building models and you're not building charts, and you're kind of flying by the seat of your pants, an OK model that is able to be accessed by more people is actually strictly better.

So when we're looking for partners, we actually have a SaaS platform. So if you go to www.einblick.ai, you can sign up and use our platform for free. And it will be free for as long as you want. You can have access to the generative AI features.

We also can support partners with education, and we can have a conversation if the SaaS model doesn't work. But whether or not your team is data heavy, as long as you have some demand for more access to data, I think Einblick is able to do that for you.

So encourage you to swing by, check us out, type in a query. But thank you very much for listening.