April 6, 2022
Data infrastructure is rapidly growing and evolving along with infrastructure for AI/ML, with the latter growing largely independent from the former. An emerging generation of AI/ML tooling emphasizes data-centric versus model-centric approaches to the ML development lifecycle. These tools recognize that data is the foundation for AI and seek to open opportunities for all data professionals to participate by eliminating the unnecessary complexity of traditional model-centric solutions.
Continual is an operational AI platform for the modern data stack powered by a declarative data-centric workflow. It enables data and analytics teams to build continually improving ML models— from customer churn to inventory forecasts — directly on their Snowflake data cloud without complex engineering. Continual’s declarative approach to ML facilitates automation of the end-to-end AI lifecycle on Snowflake from model training to inference and monitoring. By allowing any data professional with knowledge of SQL to deploy production AI use-cases, Continual radically simplifies deploying AI at scale.
Snowflake’s cloud data platform enables users to work with their data, virtually without limits on scale, performance, or flexibility. The partnership of Continual and Snowflake lets customers operationalize AI at scale to drive revenue, operate more efficiently, and build innovative products and services.
By the way, you can also learn more and see a demo of Continual on Snowflake at our recent webinar replay, available here.
Machine Learning can essentially be boiled down to three ingredients: data, code, and infrastructure. The last two generations of ML platforms have focused on code and infrastructure. These platforms have achieved remarkable benefits and made creating, training, and deploying ML better. Nevertheless, these solutions still present substantial challenges to end users. Businesses trying to adopt ML face complex pipeline and infrastructure jungles, significant time and skill requirements, and ongoing maintenance and operational burden. These hurdles are some of the reasons that many ML projects fail to launch and return business value.
Similar to how compilers, databases and operating systems evolved to unlock computing to a broader set of users, shielding them from lower-level complexity, the movement toward declarative and data-centric approaches to ML seeks to abstract away the underlying complexities best tackled by researchers and other ML specialists. With this new generation of data-centric ML platforms, data teams can operate and scale the ML lifecycle including model creation, deployment, maintenance and monitoring, all while focusing on what they know best – their business data and solving business problems.
Continual is the first operational AI platform built natively for cloud data warehouses and the modern data stack. Powered by a data-centric workflow, Continual lets everyone on the data team easily build, deploy and maintain continually improving predictive models directly on the data warehouse. There’s no additional infrastructure, complex coding or pipeline maintenance required.
Continual’s data-centric workflow incorporates:
MLOps isn’t truly ‘ops’ without support for a GitOps-friendly workflow. Data and analytics engineers have embraced tools like dbt partially because they combine simplicity with software engineering best practices. Continual’s declarative design embraces a similar philosophy and integrates seamlessly with dbt and existing CI/CD workflows. With Continual’s dbt integration, users can leverage their existing dbt models to extend dbt workflows to support operational AI with Continual.
The simplicity of the Continual AI platform could not be fully realized without Snowflake’s cloud data platform as the foundation. Snowflake combines the cloud data warehouse, the flexibility of big data platforms, the elasticity of the cloud, and live data sharing at a fraction of the cost of traditional data platforms.
A significant portion of a machine learning project is consumed by accessing and cleaning data. Disparate data sources can be difficult for data professionals to find, let alone gain access to. Snowflake is the consolidated data source where users can easily discover, access, clean and extract features from data. Continual leverages Snowflake’s efficient, dedicated virtual warehouses to query the data for building feature sets and machine learning models and then writes all model predictions back into Snowflake. Like other Snowflake workloads, only SQL knowledge is required, and compute is isolated, avoiding resource contention with other data cloud users. Having Snowflake at the core saves data teams enormous amounts of time and simplifies the ML workflow.
The Snowflake Data Exchange makes it easy to share data between teams, departments, and organizations. It’s easy and secure for data teams to enrich feature sets and improve model performance by accessing and contributing datasets without using an API, FTP, or cloud bucket storage.
Additionally, a continuously growing selection of modern data tools built around Snowflake can be incorporated for the operational data lifecycle including integration, transformation, BI, reverse ETL and event tracking.
If you’re new to Continual, try this quickstart to see how Continual + Snowflake work together. If you'd like to work through a real-word example of predicting customer churn using Continual, follow this tutorial. And again, you can also learn more and see a demo of Continual on Snowflake at our recent webinar replay, available here.
Andrew Tsao discusses the main reasons he chose to join the Continual team.
Traditional AI platforms require users to manage complex data infrastructure and write bespoke ML pipelines for each case. We let users build predictive models using a declarative workflow that radically streamlines operational AI at scale.