Snowflake has introduced the GenAI tool Cortex Analyst. With it, customers are able to perform analysis on files in their Snowflake Data Cloud environment based on normal queries. This should speed up the retrieval of information from dashboards.
The new GenAI tool should help users build dashboards for their business information or perform analysis. Whereas previously this required special help from data scientists, with the associated delays, with the use of a GenAI tool this should be faster.
Instead of data scientists having to create a dashboard or modify an existing dashboard, Cortex Analyst allows business users to directly query and analyze the data in the Snowflake Data Cloud environment.
In the process, questions asked by employees in conversational language are automatically translated into SQL queries that can then run on Snowflake.
LLMs and roadmap
Under the hood, Cortex Analyst runs on several LLMs, including Mistral AI SAS and Meta’s Llama LLMs. Users also have the option of using OpenAI’s LLMs on Azure.
The process by which Cortex Analyst uses these LLMs to process search queries consists of several steps. As a first step, the tool instructs the LLMs to generate SQL code to perform the analyses desired by the users.
Then Cortex Analyst runs the generated queries through an automated debugging model called Error Correction Agent. Among other things, this tool detects situations where an LLM tries to query data with hallucinated commands that are not part of the SQL syntax.
In the next and final step, all LLM-generated queries are run through the Synthesizer Agent component. This component distills the queries into a single code snippet that can perform the final requested analysis.
Increased reliability
To increase reliability and ensure that the LLMs used do not get confused for generating the search queries, Snowflake Cortex Analyst uses simple schemas. These are the files that define how the data set in question is formatted in an analytics project.
Cortex Analyst is configured to not process user search queries that may be too complex for the LLMs in question. In doing so, it then suggests alternative search queries that can be reliably answered.
Snowflake’s GenAI tool is now in public preview. A number of improvements to the tool will be made in the coming quarter. For example, an update that adds support for multi-turn chat interactions. Another update includes the ability to perform more complex data processing tasks.
Also read: Snowflake brings Cortex AI to regions where LLMs are not yet available