Blog

Come Learn With Us

The data landscape is complex, and applications for Artificial Intelligence and data are endless. Join us on the journey as we bring the power of data to everyone.

Informational

The Practical Importance of Clean Data in AI Systems

Discover the best practices of data cleaning and its profound impact on AI system outputs. Proper data cleaning enhances accuracy and reliability, empowering AI systems to deliver actionable insights with confidence.
Product

Analytics Agents

Our view on how AI agents are fundamentally changing the world of data analytics.
Product

Tales from the Trenches: Why Building Conversational Analytics is Harder Than You Think

This blog uncovers the hidden product requirements of developing a conversational analytics platform. From overcoming the limitations of traditional dashboards, to addressing the complexities of standardization and evolving business needs, we delve into the insights we have gained so far in our journey at Numbers Station AI.
Product

Launching Numbers Station Cloud for Business Users

With the release of Numbers Station Cloud, we bring our enterprise product to users from businesses of all sizes and enable them to discover and verify business impacting hypotheses using data through a trusted and engaging interface. 
Intern Research

A Case Study: Text-to-SQL Failures on Enterprise Data

Following our blog Text-to-SQL That Isn't, dive deeper into the complexities around real-world SQL queries in the modern enterprise with the case study by intern Manasi Ganti out of The University of Washington.

Research

Text-to-SQL That Isn't

We’ve seen incredible prototypes and demos of text-to-SQL solutions that show the dream of self-service analytics where business users, who have no knowledge of SQL or the underlying data, can ask questions to gain insights over their data. In this post, we’re going to explain why text-to-SQL is only an 80% solution and realizing self-service analytics requires moving away from the classical text-to-SQL paradigm.

Product

Programmatic Model Serving with Databricks in Python

At Numbers Station, we've partnered with Databricks to explore the cutting-edge of running foundation models on their platform. This post details our journey and the steps we took to integrate AI seamlessly into the data stack, leveraging Databricks' Python SDK.
Research

DuckDB-NSQL: How to Quack in SQL

DuckDB-NSQL, a collaboration between Numbers Station and MotherDuck, is releasing a text-to-SQL model tailored for DuckDB's unique SQL dialect. Designed for local analytics, it ensures fast, private data processing on your laptop. This model represents a leap in accessible, secure data analytics. You can find links to our GitHub repo and Hugging Face (including GFFU format) and the MotherDuck blog in this post.
Research

Three Lessons to Bring Generative AI into Enterprise Production Systems

Despite the hype around LLMs and their potential, enterprises struggle to get these models in production. In this blog, we outline the challenges presented in today's environment and share key pillars to enable enterprises to successfully bring generative AI into production to supercharge analytics.

Product

Vertically integrated AI is the answer to legacy data problems

Learn what we’ve discovered applying LLMs to data over the last decade, starting with our research at the Stanford AI Lab to now serving modern enterprises, and explore how a vertically integrated AI solution leveraging custom-built LLMs is revolutionizing data analytics. Key topics include addressing existing legacy data challenges, the strategic advantage of owning your AI and key features of Numbers Stations architecture.

Informational

The Future of Enterprise Analytics is AI

Despite technological advances, the analytics workflow remains cumbersome, mirroring processes from 40 years ago. However, AI, particularly large language models (LLMs) offer a transformative shift. Prioritizing the human experience, AI can directly provide accurate results based on user queries. At Numbers Station, the emphasis is on genuine knowledge management, underscoring that the future of analytics lies in optimizing the human experience with data.
Informational

Private vs. Third-Party Foundation Models: Navigating the Path to an AI-Driven Future

This blog delves into the choices and challenges organizations face when deciding between third-party and private FMs for enterprise applications and data analytics. It evaluates the convenience of third-party models but brings attention to significant drawbacks like vendor lock-in and IP concerns. Learn more about the benefits of private FMs for a secure AI journey in the modern enterprise.
Informational

The Practical Importance of Clean Data in AI Systems

Discover the best practices of data cleaning and its profound impact on AI system outputs. Proper data cleaning enhances accuracy and reliability, empowering AI systems to deliver actionable insights with confidence.
Product

Analytics Agents

Our view on how AI agents are fundamentally changing the world of data analytics.
Product

Tales from the Trenches: Why Building Conversational Analytics is Harder Than You Think

This blog uncovers the hidden product requirements of developing a conversational analytics platform. From overcoming the limitations of traditional dashboards, to addressing the complexities of standardization and evolving business needs, we delve into the insights we have gained so far in our journey at Numbers Station AI.
Product

Launching Numbers Station Cloud for Business Users

With the release of Numbers Station Cloud, we bring our enterprise product to users from businesses of all sizes and enable them to discover and verify business impacting hypotheses using data through a trusted and engaging interface. 
Research

Text-to-SQL That Isn't

We’ve seen incredible prototypes and demos of text-to-SQL solutions that show the dream of self-service analytics where business users, who have no knowledge of SQL or the underlying data, can ask questions to gain insights over their data. In this post, we’re going to explain why text-to-SQL is only an 80% solution and realizing self-service analytics requires moving away from the classical text-to-SQL paradigm.

Product

Programmatic Model Serving with Databricks in Python

At Numbers Station, we've partnered with Databricks to explore the cutting-edge of running foundation models on their platform. This post details our journey and the steps we took to integrate AI seamlessly into the data stack, leveraging Databricks' Python SDK.
Research

DuckDB-NSQL: How to Quack in SQL

DuckDB-NSQL, a collaboration between Numbers Station and MotherDuck, is releasing a text-to-SQL model tailored for DuckDB's unique SQL dialect. Designed for local analytics, it ensures fast, private data processing on your laptop. This model represents a leap in accessible, secure data analytics. You can find links to our GitHub repo and Hugging Face (including GFFU format) and the MotherDuck blog in this post.
Research

Three Lessons to Bring Generative AI into Enterprise Production Systems

Despite the hype around LLMs and their potential, enterprises struggle to get these models in production. In this blog, we outline the challenges presented in today's environment and share key pillars to enable enterprises to successfully bring generative AI into production to supercharge analytics.

Informational

The Future of Enterprise Analytics is AI

Despite technological advances, the analytics workflow remains cumbersome, mirroring processes from 40 years ago. However, AI, particularly large language models (LLMs) offer a transformative shift. Prioritizing the human experience, AI can directly provide accurate results based on user queries. At Numbers Station, the emphasis is on genuine knowledge management, underscoring that the future of analytics lies in optimizing the human experience with data.
Informational

Private vs. Third-Party Foundation Models: Navigating the Path to an AI-Driven Future

This blog delves into the choices and challenges organizations face when deciding between third-party and private FMs for enterprise applications and data analytics. It evaluates the convenience of third-party models but brings attention to significant drawbacks like vendor lock-in and IP concerns. Learn more about the benefits of private FMs for a secure AI journey in the modern enterprise.
Informational

Overcoming Self-Service BI Hurdles with Generative AI

Explore challenges of self-service business intelligence (BI) including the hurdles of information discovery, flexibility limitations, and consistency. Discover how tailored Generative AI solutions are revolutionizing data exploration, streamlining resources, and propelling businesses forward in an ever-evolving digital landscape.
Research

NSQL-Llama-2-7b: Advancing SQL Foundation Models for Personalized Analytics

This blog introduces NSQL-Llama-2-7b, Numbers Station's new SQL generation foundation model built off of Meta's Llama 2. NSQL-Llama-2-7b achieves up to a 15.5-point execution accuracy improvement over the previous NSQL 6b model while narrowing the performance gap with GPT-4 on standard SQL benchmarks.

Take the next step

Conversational analytics is just a question away.