By: Karthik Srinivasan Raman, Head of Snowflake Tech Consulting | LTIMindtree

Artificial intelligence (AI) is rapidly transforming enterprises, offering businesses a powerful tool to automate tasks, gain deeper insights from data, and optimize decision-making. However, deploying AI solutions often comes with substantial challenges. Businesses may lack the in-house expertise to develop and maintain complex AI models. Additionally, the traditional infrastructure required for AI development can be costly and cumbersome to manage. This creates a significant barrier for organizations eager to harness AI’s transformative potential.

For companies seeking to enhance their operations with AI, Snowflake Cortex offers a substantial advantage. Snowflake Cortex AI is a game-changer designed to democratize AI by bringing its power directly to your data, all within the familiar and secure Snowflake environment.

Introduction to Cortex AI

Data is the new oil, and AI is the refinery. In today’s hyper-connected world, businesses are drowning in data but thirsting for insights. Snowflake, a pioneer in cloud data platforms, has transformed how organizations manage and access their data. Now, Snowflake is redefining the future with Cortex AI.

Snowflake Cortex offers a managed service that streamlines the deployment of large language models for businesses. These models, integral to generative AI applications, can be utilized directly within the Snowflake platform.

With Cortex AI as your partner, you can:

• Accelerate insights: Rapidly uncover hidden patterns and trends within your data to inform strategic decisions and seize opportunities.
• Optimize operations: Automate routine tasks, streamline workflows, and improve overall operational efficiency.
• Propel growth: Make data-driven predictions and recommendations to fuel innovation and achieve unprecedented business growth.

Cortex AI helps you unlock the full potential of your data to drive business growth and innovation.

Understanding the core components of Snowflake Cortex

Snowflake Cortex is a suite of advanced AI features designed to leverage large language models (LLMs) for various applications. These models excel at understanding unstructured data, answering freeform questions, and providing intelligent assistance. The suite of Snowflake AI Features includes:

• • LLM functions: They let you use powerful pre-trained AI models (LLMs) directly in your Snowflake environment. These models handle tasks like summarizing text, analyzing sentiment, translating languages, and answering questions. You can access these features through easy-to-use functions in both SQL and Python.
  • Universal search (Now in GA): It helps you quickly find anything within your Snowflake environment, including databases, tables, data products, and documentation. It understands natural language and typos, allowing you to search using keywords or full sentences describing what you seek.
  • Snowflake Copilot (Now in GA): It acts as your AI co-pilot for data analysis within Snowflake. This LLM assistant helps you understand your data structure, write and refine SQL queries in natural language, and even optimize them for better results. It all happens securely within the Snowflake environment.
  • Document AI (Soon in GA): This feature utilizes powerful LLMs to automatically extract data from documents, including text and visual elements like logos and signatures. It can handle various document types right out of the box, such as invoices and contracts, and allows further customization for specific needs.
  • Cortex fine-tuning (Public preview): It offers a cost-effective way to improve pre-trained AI models for specific tasks. You can fine-tune these models using your own data within the secure Snowflake environment, all without the hefty cost of training a large model from scratch. This translates to better performance and faster results compared to other methods.
  • Cortex Analyst (Soon in public preview): It empowers business users to bypass complex queries. They can ask questions about data in plain English, uncover insights, and make data-driven decisions faster and independently.
  • Cortex Guard (Soon in GA): It acts as a safety net for AI within Snowflake. This LLM-powered tool identifies and flags potentially harmful content in your data, like hate speech or threats, ensuring your AI models are trustworthy and promote responsible use.
  • AI and ML studio for LLMs (Private preview): This interactive interface enables users of all technical levels to utilize AI with no-code development.

Figure 1: Snowflake Cortex Architecture

Snowflake Cortex AI, Snowflake: https://www.snowflake.com/en/data-cloud/cortex/

Building AI applications with Cortex AI

Potential Use Cases

Cortex AI transcends theoretical concepts and equips you with the tools to tackle a variety of business challenges. It transforms the way businesses operate and make decisions. With Cortex AI, organizations can carry out:

• Customer analytics
  • Customer churn prediction: Utilize advanced analytics to study user behavior and predict which customers are likely to churn. You can implement targeted retention strategies to enhance loyalty and reduce turnover by identifying these high-risk customers.
  • Sentiment analysis: Analyze customer feedback from multiple sources, including reviews, social media posts, and surveys. This insight helps in understanding customer sentiments, spotting emerging trends, addressing core concerns, and refining your offerings to meet customer needs better.
  • Intelligent recommendations: Build personalized recommendation systems leveraging LLMs. Analyze customer data and preferences to suggest relevant products or services, driving sales and satisfaction.

• Fraud detection
  • Analyze transaction patterns and identify deviations from normal behavior using anomaly detection powered by LLMs. This capability helps in identifying potentially fraudulent activities quickly, allowing businesses to take immediate action and safeguard their financial assets.

• Marketing and sales
  • Next-gen chatbots: Create advanced chatbots that excel in answering customer inquiries, providing support, and automating various tasks. These chatbots use large language models to comprehend natural language, offering a more personalized and engaging customer interaction experience.

• Business operations
  • Document automation: Automate data extraction from diverse documents (invoices, contracts, etc.) using Document AI. This eliminates manual data entry and improves efficiency.

• Data exploration and analysis
  • Universal Search: Find anything quickly within your Snowflake environment using keywords or natural language descriptions.
  • Snowflake Copilot: Understand your data structure, write and refine SQL queries in natural language, and optimize them for better results with the help of this AI assistant.

Cost considerations for Snowflake Cortex LLM

Understanding the cost implications when using Snowflake Cortex’s LLM functions is essential for managing your budget effectively. Here is a breakdown of the core cost drivers and how to estimate your expenses.

Core cost driver: Tokens processed

Snowflake Cortex LLM functions incur compute costs based on the number of tokens processed. A token represents the smallest unit of text handled by these functions, roughly equivalent to four characters. The raw input or output text’s equivalence to tokens may vary by model.

Understanding consumption costs

• Compute cost: Each LLM function incurs a compute cost based on the number of tokens processed, measured per million tokens. The documentation provides the details of the cost per million tokens for each LLM function and model combination. This is crucial for estimating the cost of your LLM workloads.
• Function impact: Costs vary depending on the type of function you use:
  • Text-generation functions: Functions such as COMPLETE, SUMMARIZE, and TRANSLATE generate new text and incur input and output token costs. This means the more text these functions produce, the higher the cost.
  • Text analysis functions: Functions like EXTRACT_ANSWER and SENTIMENT analyze existing text and only incur costs for input tokens. These functions do not generate new text, so the cost is associated solely with the amount of text being analyzed.

Warehouse size and cost

• Recommendation: Snowflake recommends using a small warehouse (no larger than MEDIUM) when executing queries that call LLM functions. While larger warehouses may not improve performance, their associated costs will still apply.
• Focus on compute cost: The primary cost consideration for LLM functions is compute cost based on token processing, not warehouse size.

Additional considerations

• Model selection: Different LLM models may have varying processing costs per token. The consumption table will help to identify the most cost-effective model for your needs.
• Data volume: The volume of data processed by LLM functions also affects the overall cost. To optimize expenses, consider refining your data pipelines to reduce unnecessary processing and minimize the total data handled by the LLM functions.

Best practices for cost optimization

• • Start small: Begin with smaller datasets and workloads to gauge costs before scaling up.
  • Optimize queries: Structure your queries to minimize the number of tokens processed by LLM functions.
  • Utilize available resources: Use Snowflake’s documentation, consumption tables, and code examples to make informed decisions about using LLM functions and selecting models.

• Monitor and track costs: Use Snowflake’s cost monitoring tools to track your LLM function usage and identify areas for potential optimization.

Conclusion

In a nutshell, Snowflake Cortex AI is poised to transform the data landscape by integrating advanced AI capabilities directly into the Snowflake environment. This integration allows businesses of all sizes to harness the full potential of their data with greater ease and efficiency. From automating tasks and uncovering hidden insights to building intelligent applications, Cortex breaks down the barriers to AI adoption. With its intuitive interface, cost-effective structure, and focus on responsible AI practices, Cortex democratizes AI, making it accessible to a broader range of users than ever before. As businesses continue to navigate the ever-growing data landscape, Snowflake Cortex AI positions itself as the key to unlocking valuable insights and driving data-driven decision-making, paving the way for a future fuelled by intelligent data exploration.