4 min Analytics

Snowflake lowers the barrier for building AI apps

Insight: Analytics

Snowflake lowers the barrier for building AI apps

Organizations are looking for AI-enabled applications to create smarter software. Yet, it can be difficult actually to connect enterprise data. You want to because feeding AI in this way gives the business user the most complete output possible. Snowflake is jumping into this gap at the Data Cloud Summit 2024 conference through updates to Cortex.

With the updates, Snowflake is positioning one of its key Data Cloud assets more prominently. In late 2023, the company unveiled Cortex, a fully managed service that enables organizations to build AI apps more easily. Cortex accomplishes this with features to add large language models (LLMs), task-specific models and vector search options to software. In this way, the Data Cloud can do more for AI. Snowflake originally supported mostly analytics workloads, something AI is too, but with the explosion of artificial intelligence over the past 18 months, companies are looking for new capabilities. The image below paints a picture of what Snowflake is doing with Cortex.

Cortex is now generally available after a private preview of just over six months. So, at the Data Cloud Summit, Snowflake is going full steam ahead to update Cortex and tout what the service can do. Especially since Cortex represents Snowflake’s larger goal: to reach every user in an organization. The ambition is to make the Data Cloud a place where even non-data professionals can interact to realize AI apps.

New chat experience

Snowflake will transform this ambition by introducing Cortex Analyst and Cortex Search. The bottom line is that the most commonly used LLM applications, chatbots, need to become stronger in providing the right answer. This by improving integration with business data so that a chatbot can handle structured and unstructured data. Cortex Analyst, built with the Meta Llama 3 and Mistral Large models, allows applications to be built securely on top of analytics data in Snowflake. Cortex Search, in turn, uses retrieval and ranking technology from Neeva, the company that Snowflake acquired in May 2023, combined with Arctic-embed. This makes building software that takes documents and other text-based datasets possible. For this, it relies on hybrid search, a combination of vector and text.

We already collected user reactions on presenting these two new Cortex features, which will soon go into public preview. For example, Mukesh Dubey, Product Owner Data Platform at Bayer, informed Techzine, “What if internal functional users could ask specific questions directly on their enterprise data and get responses back with basic visualizations? The core of this capability is high-quality responses to a natural language query on structured data, used in an operationally sustainable way. This is exactly what Snowlake Cortex Analyst enables for us. What I’m most excited about is we’re just getting started, and we’re looking forward to unlocking more value with Snowflake Cortex AI.”

Above, we briefly mentioned that Cortex also puts security at the heart of application development. To reinforce this issue, Snowflake will soon make Cortex Guard available. This component uses the input-output safeguard Meta Llama Guard to filter and flag harmful content in enterprise data, which may include violence, hate, or criminal activity. Cortex Guard is, therefore, primarily a feature for improving trust in AI, aka trustworthy AI.

No-code AI development

Where Cortex Analyst and Cortex Search are undoubtedly going to help is in lowering the threshold of development. Many companies also see the possibility of achieving that simplification with no-code. That’s where Snowflake is stepping further with the new AI & ML Studio. Companies can use this environment to build AI apps quickly via building blocks. A visual component, like a building block, features prescribed code for performing actions. The development process can theoretically happen much faster by building AI apps via drag-and-drop. We don’t yet know how that works out in practice. During the Data Cloud Summit 2024, the feature was still in private preview. Snowflake does promise that users can easily test and evaluate models within the studio to find the most cost-effective option for the use case.

Through the AI & ML Studio, companies will also soon have access to Cortex Fine-Tuning. This feature should help boost an LLM’s performance and deliver a personalized experience. This serverless customization features a subset of Meta and Mistral AI models. The models can be used via a Cortex AI feature, with access managed via role-based access controls from Snowflake.

Techzine is attending the Data Cloud Summit this week. Previously, we brought news of the collaboration with Nvidia and the new data catalog Polaris.