Utilizing your Snowflake data in AI and ML applications while keeping the data secure can be challenging. Even if it’s low, there’s always some chance of your data being compromised when it’s taken out of your Snowflake account.
That’s exactly why the Snowflake AI Data Cloud created Native Apps, which are applications that run directly in your own private Snowflake bubble. With this new application format, companies are publishing apps for Artificial Intelligence (AI) and Machine Learning (ML) models.
In this blog, we’ll explain Native Apps and which ones we feel are some of the strongest out there right now. We’ll also explore some great AI and ML datasets from the Snowflake Marketplace.
What are Snowflake Native Apps?
Native Apps are a relatively new feature from Snowflake where companies can build applications directly in Snowflake for seamless use with the platform’s core functionality. They are meant to provide users with new tools they can use in their account for data analysis, visualizations, or even to check if their account is running as efficiently as possible.
Apps can be either free to use, offer a free trial, or are a completely paid model. The paid model can be anything from a one time fee to a monthly subscription. Some apps are only available in certain cloud regions or for particular editions of Snowflake. These are easily identifiable as the button to install the app will show “Request” instead of “Get” if it is not currently available for you.
Tools designed to enhance a business’s AI and ML capabilities are now being developed as Native Apps, allowing them to run directly on your data without leaving your account. Let’s explore some of the best AI and ML apps we’ve seen so far.
What Are the Best Native Apps for AI & ML?
LandingLens by LandingAI
LandingLens is the industry-leading AI Computer Vision software platform from the people at LandingAI. It is an end-to-end software platform designed for domain professionals and AI experts to quickly develop and deploy visual classification systems. LandingLens’ best feature is that it allows you to create and test computer vision AI projects in minutes with little or no code, making it an accessible and intuitive platform for users.
To begin building a custom computer vision model, you do not need complex programming or AI knowledge. And now, LandingLens is available as a Native App for even faster setup in your Snowflake account.
By using LandingLens as a Native App, you get the added benefits of:
Loading images directly from Snowflake to LandingLens so they never leave the security of Snowflake.
Managed access to the app using Snowflake’s RBAC model.
Utilization of Snowflake’s other AI applications, such as Cortex Copilot, to get additional insights.
Learn more about LandingLens
To see this Native App in action, we highly recommend checking out our recent blog that walks through how to train and deploy a model in 15 minutes using LandingAI. This blog makes a compelling case that building enterprise-grade computer vision (CV) applications is easier than ever.
Kumo
Running AI and ML applications on your data can require extensive work to prepare it for ingestion into models. There is so much preprocessing to be done, which takes hours, and often, the quality of the model’s output is severely dependent on the quality of the data being input into it. Kumo has developed an application to try to alleviate these issues.
By converting your Snowflake tables into a graph, Kumo can use graph machine learning directly on raw tables, eliminating the need for much of the preprocessing. Kumo can then provide detailed model evaluation, model explainability, and MLOps to build trust and monitor models in production including explanations down to the raw data.
By using Kumo you can:
Gain insights into how various columns influence every prediction.
Understand how specific inputs contribute to individual predictions.
Uncover relevant trends hidden in raw input data.
Leverage Kumo as a Snowflake Native App to write predictions back to Snowflake tables.
Utilize Cortex to enhance predictions by incorporating unstructured data.
H2O Driverless-AI
As mentioned earlier, preprocessing your data for use in models can be challenging. When new datasets are added, it means preparing the data, combining it with the old data, and scoring the data for use in models. H2O created a Native Application that can automate that entire process, making your data ready for a model with minimal human interaction.
The whole process can be done using SQL by a data engineer, allowing them to manage the entire end-to-end process all while the data stays safe and secure within Snowflake.
We’ve had a few clients use H2O to achieve:
Automate feature engineering and model tuning
Achieve higher accuracy of predictions
Improve model transparency and explanation
What Are the Best Snowflake Marketplace Datasets for AI & ML?
Applications aren’t the only thing available within the Snowflake Marketplace. For those running AI models with Snowflake, it’s worth mentioning that datasets are also available that may enhance your models.
LandingAI Sample Data
To get started with LandLens, LandingAI offers a sample dataset so you can test the software and see how proper images should be fed into the system. Having a sample dataset can help you get used to labeling and segmenting your images to get the most out of LandingLens.
For example, the following dataset contains images of metal casting in a manufacturing process. It includes images where the casting was done perfectly and where there were imperfections. Using this data, you can label images with issues, train the model, and use the rest of the data to test the model without uploading your own data. This will help your team get a feel for just how easy using LandingLens can be.
Financial Datasets
Augmenting your in-house data with various financial and economic factors as features can provide better predictions in your model. For example, this dataset on the marketplace is free and contains many metrics such as GDP, unemployment, banking deposits, and more. This can be especially useful for predicting overall sales due to economic factors, analyzing geographic areas that may be a good location for expansion, or assessing credit risk more effectively.
How to Get Marketplace Datasets
Fortunately, these and all the other datasets on the Marketplace will not use up any space in your account when you want to utilize them. All Marketplace data is available through Data Shares. This means that no data is copied into your account, so you don’t have to pay any storage costs; you can just compute costs when querying the data. Data Shares only allow for reading the data though, so cloning the share or making any changes to it is off the table.
Closing
Snowflake Native Apps offer a powerful solution for leveraging AI and ML applications without compromising data security. By running directly within your Snowflake environment, these apps provide a secure and efficient way to access and analyze your data. As the Native App market matures, more AI and ML applications will become available.
Have any questions?
The Snowflake experts at phData can help! Reach out to us today for Snowflake guidance, insights, and support.
FAQs
Can I create my own Native App for others to use?
Yes, you can create your own Native App to be put on the Marketplace. Snowflake has a quick start tutorial for getting your first app built and published. Apps can be written with a Streamlit app, stored procedures, and functions written using the Snowpark API, JavaScript, and SQL. Snowflake even includes a testing environment to make sure it works before sending it off for the world to install.
How much do Native Apps cost?
It depends on the application. phData’s Advisor Tool app is free for instance, whereas some have a one time or a monthly subscription to use.
