Name: RapidCanvas
Uploaded: 2025-02-19T10:22:43.783Z
Duration: 2 min 33 s
Description: An AI platform combining AI agents and human expertise to deliver custom enterprise AI solutions swiftly.

rapidcanvas

An AI platform combining AI agents and human expertise to deliver custom enterprise AI solutions swiftly.

RapidCanvas

I'm going to walk you through a quick end-to-end demo of RapidCanvas where we're going to build a machine learning pipeline to predict loan default. 

So the first thing I do is create a project. Uh this isolates the use case from other use cases in the environment. Uh I'm able to collaborate and share with other users and have them work on independently, but it sets up a nice clean canvas. 

What we're first thing we're going to do is add data. Um I'm going to import data. I can pull it in from GCP Snowflake, one of 584 different connections, or I can just upload it. Just upload it from a local disk. In this case, I'm going to connect to Snowflake and actually I'm going to build the pipeline such that it reads from Snowflake. 

Um go on default. Uh reads from Snowflake and writes back to Snowflake. So pull in the metadata and put in the SQL statement and give the a test. Okay we can edit this query at a later time. We could put limit statements on it, joins and so on, filters. 

Uh but as it is, I'm going to pull the whole table in and use that as the basis for my original initial design. Okay so now I have a data set on the canvas. It's linked to Snowflake. If I run this job in the future, the latest data set will be pulled from the Snowflake database. 

Um I can perform a number of different operations here. Uh first one I'm going to do is have a look at the data [Music] itself. Okay so you can see the size and shape of the data here. 5,000 rows, 27 columns, correlation matrix. I can see that you know annual.

Income has uh positive relationship with loan amount. You lend me more money the more I earn. It's got a negative impact on interest rate. Higher more I earn the lower my interest rate and so on. So you can see the relationship between the parameters here. Um and I've also got a set of alerts. You know you can see things like uh there's a number of people who don't have mortgages. No real surprise there. But uh this one I might want to dig into. This annual income is highly skewed. It seems like that's quite a big amount of skew you 48 times. 

So um I can jump into advanced analytics and this pops out another tab. So I can leave this open and refer back to it. It shows me the alerts and so on. But here in the variables I can find annual income and I can see that it goes from you know zero right up to 6 million. So there's definitely some skew in that data. I can go and look exactly how that looks with extreme values. So I say the minimum reported income is $7,500. The maximum is 6 million. So you can see there's a few above 400,000. That's definitely something I might want to fix before I build a machine learning model on this data. 

So let's go and do that. Let's go back to the canvas and now we're going to clean up this data and get ready to model. So I'm going to use something called an AI assisted recipe. This pulls in the data and gives me a conversational interface on this data. So in this case I'm going to do a very simple operation. I'm going to remove rows salary sorry income is greater than 3991 2 3. So I'll get

Rid of 4,000 400,000 up what's happening now is it sends this prompt and the schema of the data to a large language model. The large language model generates code which we test validate and if it looks okay we run it. 

So just to show you the difference there are 5,000 rows we said in the source data in this intermediate data set I'm now down to 4,992. So I've lost eight rows. Those eight rows with data above 400,000. Um other types of operations if you look at the data maybe I want to clean up you know there's an employee length field here where I've got some kind of categorical variable it's text and I want to change it to numeric. 

So I want to replace the m length column with numeric only values. Okay it's sending the prompt and schema getting code. The because being returned incl M length is now numeric. So that might have taken me a little while with regular expressions in Excel and that couple of seconds I generated code ran the tested the code and ran the code. 

I can add this to the recipe. What this does is drop back on the canvas. Okay so data and in doing so it's built a code recipe which is doing both of those transforms. So if I go into the code you can see the function this recipe is now converted into code. 

Every time I run it we'll do exactly the same thing. It can be audited. There's no interaction with the llm. There's no additional cost the next time I run it so it's fixed. Um in here I can do more than just data transformation. 

So I can create new data sets but I can also ask questions create create.

Charts get prompt suggestions or build a model. So if I just ask a question, what is the average amount? My bad spelling or Str typing will be taken care of by the llm. It gives me an answer of $1,166.53 is the average. Um, if I want to plot um, uh loan amount versus annual income versus interest rate, again send the prompt and the schema to the llm. I get code back, python code back that I'm going to execute in the platform. I'm going to validate, execute, uh, and then produce the result.

So here I've got an interesting chart which is showing me um, annual income on the bottom, loan amount on the left. You can see that there's a nice line up here and a line along the top which suggests the set maximum for my loan is 35,000. The typical limit on a loan is, you know, 50% of your salary. So there's a few higher than that and I can zoom in and take a little look at those and see there's one over here, loan 25k on a salary of 30k. So it's a little out of work.

Okay, um, I can can add. Let me just know this is an interactive chart. Let me just add it to the recipe. Again, this is going to drop it onto the canvas, run it, and come out. You'll see the recipe runs and it's going to generate uh, the data set that I saved and this chart. So there's the clean data and there's the chart. Just validate the charts. Okay, you know, as I say, it's interactive. I can zoom in on a slice, I can zoom in on a section, go back to the start. Um, and I now have a data data set where I've removed the skew in the annual income and I have set employee length.

To be um a numeric value. I'm now going to build a predictive model and I have in this data set loan status which tells me whether or not this loan was fully paid off or whether it was went into default. So charged off it's telling it's in default. 

So I'm going to use that to build a model. If I didn't know what model to build or what to do with this data anywhere I see this orange AI guide I'm able to go and ask uh questions. You know what kind of model could I build on this data set uh and it'll send this uh information to the llm. The llm will give me a much more broad open answer than previously where it's telling me is doing specific operations on the data. 

Here it's looking at the data and using the LM to give me some suggestions. So classification on loan status is the the first thing it's telling me to do. So that let's go and do that back to the canvas. I get to add to this data set this rapid model recipe. 

Um I specify the target which is status and it determines it's binary classifier. You can see in here let me just remove that just let's leave it a normally. So it's a normally detection binary classification clustering multiclass and regression but I can also do time series forecasting. 

Okay so I'm going to save that and run it output we can see uh it's an output data set. This is the input data set with a couple of columns added at the end. So this is the prediction that we we've added so this is created by the model and it's the probability of it being paid off versus going into default. 

So we've added three.

Columns to the data. Um this was done by a prediction. Uh from the model I can actually make that prediction available to service def three just three R service. This is just choosing which environment is going to run in. I can make those different sizes and shapes and it adds a rest end point where I can which I can call for my application. 

Okay. Um I've got other information there but we'll come back to it. So what I've got now is a an ENT in pipeline reading from Snowflake cleaning and prepping the data creating a chart building this. Uh finary classification model generating output. I can run another code assist recipe and here I'm going to take um probably the date of the loan so issue D and I think there an address field in there so I can match it back. 

And then I'm going to take these last three columns. So I want to extract uh issue date address for to spell address. Uh address the prediction it was for the probability of pully paid or Char off. Okay again this sends this prompt asking me to extract those columns and I get 4992 rows only five columns add that to the recipe. 

Uh predictions save and run and as it runs on the output it's going to generate a data set that I want to write back to snowflake. To do that I can come here and specify snowflake connector table name is default prediction database was lab schema was public Warehouse unroll. Okay and I can choose to append or replace. I'm going to append and I can now save and export. So that's going to export this file out back to snowflake so wherever I run.

This pipeline if I run the pipeline up here it'll pull the data score the data and write it back to the prediction service. I can also schedule those jobs um on periodic basis so daily weekly or in whatever interval I choose and have those run automatically. 

The other thing I'm able to do so those are job we talked about scenarios uh settings information there's the ability to create a data app. So I'm going to create a new data app on the model I'm going to call that uh explore prediction of default and when I create that it's going to take a few seconds to run. 

So while I'm waiting for that to launch I'll actually add a second just to show the other capabilities of the platform. I can create a data app that allows me to provide a Q&A a chatbot Q&A on top of the data on the canvas. So the data I've created any file set I've created on the canvas I can upload an Excel or a CSV and ask questions of it. 

I can connect to snowflake or my SQL or some other data SQL data source and have a chat with the data in that database. I can upload PDFs and add PDFs so let's do that or the prediction service which I've got buil in the background. 

So this is uh CL application requirements PDF and when I create that when I go and run it I can upload the PDF and then start chatting with that PDF. Okay the these apps take a few minutes to load one is the explore the data app so I said earlier I was going to skip the model output here I've got the same model output but with explanation. 

So here I can see the feature important auction.

Basically um more interestingly I have the ability to run what um the first time I run this it takes a few seconds to start the service up in the back end. 

Okay just see what happens if I bump that interest rate out buttons. If we double the interest rate and give them an interest rate 20% again it sends the data in with this change and I can see the change in probability that interest rate do there. 

So it's dropped it by to predict for I have the ability to do ad hoc predictions. I can pull a file in score it and download it and I can give a business user that same ask AI interface where they can create charts and ask questions of the data. 

Note, that there may be slight differences in text and video. Watch video for the original content.

RapidCanvas Video