New Technology / Ai Development
Track AI development, model progress, product releases, infrastructure shifts and strategic technology signals across the artificial intelligence sector.
Snowflake VP of AI Baris Gultekin on Bringing AI to Data, Agent Design, Text-2-SQL, RAG & More
Topic
AI in Enterprise Data Management
Key insights
- Bearish Gultiken is the Vice President of AI at Snowflake
- Snowflake describes itself as the AI data cloud
- Snowflake acquired Niva, an AI-powered web and personal knowledge-based search engine, in May 2023
- Snowflakes core philosophy is to bring AI to the data rather than sending sensitive data out to model providers
- The ongoing unlock of unstructured data is making 80-90% of enterprise information previously trapped in PDFs and other documents queryable
- AI coding assistants are changing product management by enabling rapid prototyping of working features
Perspectives
short
Baris Gultekin (Snowflake)
- Highlights the importance of bringing AI to data rather than sending data to model providers
- Claims that AI enables the extraction of structure from unstructured data, increasing usability
- Proposes that recent advancements in reasoning models have improved text-to-SQL applications
- Argues that modularizing AI skills enhances operational efficiency
- Emphasizes the need for enterprises to rapidly adopt AI to gain competitive advantages
- Warns that the integration of AI capabilities is transforming product management and development
Nathan (Interviewer)
- Questions the effectiveness of AI models in navigating personal data compared to human intuition
- Challenges the notion that enterprises can easily switch between AI models without significant costs
- Highlights concerns about the potential commoditization of AI models and their impact on businesses
- Raises doubts about the ability of AI to fully understand and utilize proprietary enterprise data
Neutral / Shared
- Notes that the integration of AI into enterprise systems is reshaping data management practices
- Observes that the competition among AI model providers is benefiting consumers and businesses
- Mentions the importance of security and governance in AI deployment within enterprises
Metrics
unlocked_data_percentage
80-90%
percentage of enterprise information that is now queryable
This significant percentage indicates a major shift in data accessibility for enterprises, enhancing decision-making capabilities.
the massive ongoing unlock of unstructured data, which is making the 80-90% of enterprise information that was previously trapped in PDFs and other documents, queryable
unstructured_data_percentage
80 to 90%
proportion of all data that is unstructured
Understanding the volume of unstructured data helps organizations prioritize data processing strategies.
80 to 90% of all data is unstructured data.
growth_rate
fastest growing product
refers to the demand for Snowflake Intelligence
Indicates strong market acceptance and potential for revenue growth.
it's been used by business users. And it is the fastest growing product that we have on snowflake
software_development_speed
increasing dramatically
The pace of software development in the market.
A faster development pace can lead to more rapid innovation and competitive dynamics.
the pace of software development is increasing dramatically.
help_desk_automation
50%
percentage reduction in help desk tickets
This significant reduction can lead to increased efficiency and cost savings for IT departments.
you can cut help desk tickets by more than 50%
automation_guarantee_time
week four
time frame for achieving help desk automation
A clear timeline for automation helps organizations plan and allocate resources effectively.
guarantee 50% help desk automation by week four of your free pilot
cost
higher cost certainly inferensely wise but definitely a lot lower cost in terms of AI engineering time USD
cost comparison between large language models and specialized models
Understanding cost implications helps businesses choose the right AI solutions.
higher cost certainly inferensely wise but definitely a lot lower cost in terms of AI engineering time
size
multiple orders of magnitude smaller dimensionless
size comparison of Snowflake's document extraction model to large language models
Smaller models can lead to faster processing and lower costs.
It is in multiple orders of magnitude smaller than these large language models.
Key entities
Timeline highlights
00:00–05:00
The integration of AI capabilities at Snowflake enables enterprises to unlock previously inaccessible data, enhancing data analysis and operational efficiency.
- Bearish Gultiken is the Vice President of AI at Snowflake
- Snowflake describes itself as the AI data cloud
- Snowflake acquired Niva, an AI-powered web and personal knowledge-based search engine, in May 2023
- Snowflakes core philosophy is to bring AI to the data rather than sending sensitive data out to model providers
- The ongoing unlock of unstructured data is making 80-90% of enterprise information previously trapped in PDFs and other documents queryable
- AI coding assistants are changing product management by enabling rapid prototyping of working features
05:00–10:00
AI enables the extraction of structure from unstructured data, increasing its usability and allowing for more comprehensive data analysis.
- Customers have been using Snowflake Inc. mostly for structured data initially
- The platform helps break data silos to run analysis across large amounts of data
- AI unlocks unstructured data, allowing extraction of structure from documents
- Users can easily analyze contracts and find specific information about them
- Pipelines for data classification and extraction have become simpler with AI
- Natural language interfaces are being developed to democratize access to data
10:00–15:00
Recent advancements in reasoning models have improved the quality of text to SQL applications, enabling business users to access data directly, which enhances operational efficiency.
- Models have traditionally struggled with text to SQL due to high expectations of quality and low margin of error
- Understanding semantics is crucial for accurately retrieving data, as definitions can vary
- Recent advancements in reasoning models have improved the quality of text to SQL applications
- Snowflake Intelligence is a product that enables business users to access data without needing an analyst
- The product has seen tremendous demand and is the fastest growing offering from Snowflake
- AI can assist in building semantic models by utilizing data, metadata, and historical queries
15:00–20:00
The increasing pace of software development is leading to a breakdown of traditional silos, fostering competition and innovation among companies.
- Active development with all parties in the open semantic interchange to define the interface for open interchange of the semantic model
- Customers can build out their semantic model in Snowflake and reuse it in other places that support open interchange
- The pace of software development is increasing dramatically
- There is talk of companies deleting systems of record and rolling their own in-house solutions
- Companies that used to be partners may start to compete for customer territory
- The silos are coming down, benefiting customers and consumers
20:00–25:00
Server automates help desk tasks, reducing ticket handling time by over 50%, allowing IT teams to focus on more productive work.
- Server can cut help desk tickets by more than 50%
- Server allows IT teams to describe needs in plain English and writes automations in seconds
- Manual provisioning can take a week or more before actual work can start
- Server guarantees 50% help desk automation by week four of a free pilot
- Quality in rag solutions is determined by the quality of the embedding model used
- Handling complex documents like PDFs is increasingly important for extracting information accurately
25:00–30:00
Cloud 45 opus and Gemini 3 provide cost-effective solutions for document understanding, enabling efficient processing of large datasets with specialized models.
- Cloud 45 opus or Gemini 3 solved the problem off the shelf in terms of understanding documents
- Different use cases call for different tactics and models
- Processing hundreds of millions of documents is not feasible with cloud due to cost and throughput
- Snowflake has a document extraction model that is smaller, cheaper, and faster than large language models
- Custom models make sense for customers with large amounts of data and specific throughput or cost requirements
- A well-tuned rag solution with a large language model is usually the go-to scenario