Blueprint and governance
The first use cases, the semantic model for Cortex Analyst, the allowed models, the cross-region setting and the roles that may call Cortex are agreed with the data owner, security and the analytics lead before build.
Platform implementation
Snowflake Cortex
We implement Snowflake Cortex so analysts and applications can use language models on governed data without moving it out of Snowflake, starting with Cortex Analyst, Cortex Search and AI functions in SQL.
Cortex brings the models to the data, which suits organisations that have already put their governance into Snowflake roles and policies. Not every model is offered in every Snowflake region, and when one is missing, cross-region inference has to be switched on at account level, which sends data to another region, so residency teams in the UAE and Saudi Arabia should take that decision explicitly.
Cortex Analyst answers business questions in plain language from a semantic model or semantic view that we build and test with the metric owners. Cortex Search indexes documents and text columns for retrieval, AI functions such as AI_COMPLETE and AI_CLASSIFY run inside SQL pipelines, and Document AI pulls fields from invoices and forms. Access is granted through the CORTEX_USER database role alongside normal Snowflake privileges.
Snowflake Cortex and related product names may be trademarks of their respective owners.
How we implement Snowflake Cortex
Implementation, customisation, deployment and ongoing support, with governance your steering committee can follow.
Verified questions
Real questions with checked answers added to the semantic model, so the SQL that Cortex Analyst writes is tested before users rely on it.
Document processing
Invoices, contracts or forms turned into tables, with low-confidence fields sent for human review before the data is used.
Cost control for AI functions
Credit use by function and by warehouse tracked from account usage views, with limits on who can run large batch jobs.
Hypercare and managed support
Snowflake adds and retires Cortex models and features through its weekly releases, so we review changes, rerun question tests on the semantic model, and watch credit use, failed calls and user feedback on answers.
When Cortex is enough
Cortex suits organisations whose data and governance already live in Snowflake and that want AI on that data without running a separate AI platform or moving data out.
- Core data already in Snowflake, with roles and masking in place
- Metric definitions ready for a semantic model
- A decision on cross-region inference
- A credit budget for AI functions agreed with finance
Plan Snowflake Cortex with Hosn Al Tomoh
Tell us what the platform needs to do and when. We will propose a first phase with scope, team and an itemised quotation.