Numbers Station is revolutionizing data analytics with Agents 🌐

CV Deep Dive

Today, we’re talking with Chris Aberger, Co-Founder and CEO of Numbers Station AI.

Numbers Station revolutionizes data analytics by leveraging the power of LLMs across the entire data stack to increase the effectiveness of data teams and its business users. Founded by Chris Aberger, Ines Chami, Sen Wu, and Chris Re in 2021, Numbers Station marries structured data with AI, reducing the burden on data teams by powering action-driven conversational analytics and enabling more efficient data workflows. The company’s mission, as explained by Chris, is to streamline data analytics processes and allow technical & non-technical users to interact with their data more intuitively and productively.

Today, Numbers Station supports a variety of use-cases including internal data team acceleration and white label applications, making it easier for organizations of all sizes to derive insights and take action from their structured data. Notable customers include enterprises in financial services, retail, and commercial real estate. In early 2023, the company announced a $17.5m Series A funding round led by Madrona, with participation from Norwest Venture Partners, Factory, and notable angel investors such as Cloudera Co-founder Jeff Hammerbacher.

In this conversation, Chris takes us through the founding story of Numbers Station, his vision for integrating AI with structured data, and Numbers Station’s roadmap for the next 12 months.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Chris 💬

Chris - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to found Numbers Station? 

Hey there! My name is Chris Aberger, co-founder and CEO of Numbers Station. My journey to building Numbers Station started with my PhD at Stanford, where I focused on a mix of data and AI systems. I began with hardcore databases and gradually transitioned more into AI systems towards the end of my PhD. After that, I spent four years at another startup called SambaNova Systems, leading the machine learning team and doing a lot of foundation model training.

Through my work at SambaNova and during my PhD, I got really excited about foundation models and large language models. This was pre-ChatGPT, so it was a bit before it became the trendy thing, but I saw the trend coming and was excited about applying it to problems in the data space. I saw a big opportunity in combining structured data and AI—data and analytics teams spend a lot of time on tedious tasks that people don't want to do, like cleaning data, prepping it for reporting, and issuing SQL queries. These tasks are perfect candidates for acceleration or automation using large language models or foundation models. 

This eventually led to the creation of Numbers Station. 

Give us a top level overview of Numbers Station—how would you describe your company to those who are perhaps less familiar with you? 

The idea behind Numbers Station is to ease the burden on data teams by automating repeatable tasks and increasing the efficiency of data teams. 

Tactically, we focus on conversational analytics—allowing users to have conversations over their structured data to gain new insights. That said, our platform goes beyond just conversational analytics to more agentic functions, like taking actions to address findings from the data. 

Ultimately, you want to do something with the insights you gain, and LLMs can help automate some of those actions. 

We also handle more than just descriptive analytics, extending into predictive analytics and causality. Our platform is designed to be a foundational tool for data analytics, aimed at relieving data teams from mundane tasks and enabling them to focus on more strategic, high-level work.

You’ve mentioned the term ‘multi-agent architecture’ in reference to Numbers Station’s products. Could you dig into that a little bit, and walk through the architectures you’ve chosen to use? 

There are a bunch of different agents we've built out. When we talk about multi-agent architecture, it's similar to other agentic platforms you might have seen, like those built on LangChain. However, most of those platforms aren't centered around structured data, which is what organizations usually deal with. We've built a lot of infrastructure around these structured databases and data.

It's interesting and challenging because LLMs are suited for unstructured data, which is how they're trained. Structured data is a different ballgame, and the infrastructure around the models is crucial. In structured data analytics, accuracy is paramount. Getting something 95% correct is still 100% wrong if it needs to be 100% right.

Our platform has extensive infrastructure around RAG (Retrieval-Augmented Generation) and knowledge curation. This architecture involves a network of various agents working on top of structured data. We have agents that curate metrics or clean data, others handle intent, and some translate queries for databases. We also have tool agents and downstream task agents that build on top of this infrastructure. That's the multi-agent architecture we're developing.

Have there been any specific sectors or verticals where you've found the most exciting use-cases for Numbers Station? Where are your users finding the most value in the product? 

Financial services have always been great early adopters of cutting-edge technology. They deal with tons of structured data and have interesting use cases. Retail is also a good fit, with large amounts of structured data to derive insights and actions from. We've also had success in commercial real estate.

One interesting aspect of our technology is its general platform nature. It's not industry-specific but has industry-specific knowledge components built on top. Early on, it was about figuring out where this would land first. So far, financial services, retail, and commercial real estate have been the three main areas where we've had success.

Let’s walk through the products that Numbers Station has - what should new customers know when they're first interested in incorporating Numbers Station into their AI stack? 

We have deployment methods for both Cloud and Enterprise. If you're in the cloud on AWS, Azure, or Google Cloud, we can deploy at a lower cost compared to going into your VPC. For use cases, we often start with conversational analytics, helping organizations by accelerating their data teams. 

There are two scenarios: internal and white label style. Internally, we sell the platform directly to data teams to make their work faster and more efficient. In white label use cases, organizations use Numbers Station to power the products they sell to their customers. Our goal is to democratize data access, which helps our users gain insights without dealing with mundane, repeatable tasks.

We've seen an explosion of interest in AI agents and agentic frameworks in 2024. It seems like you front-ran some of the extreme interest people are taking in them now - what was your thinking initially? 

I'm smiling because if you look at the original pitch of this company from our first seed deck, the high-level vision has stayed the same, though some names have changed. We've always aimed to get insights from data and automate tasks. Initially, it was more like RPA-style things before agents or LLMs became popular. It's amazing how the market has evolved to align with our vision. 

We've always had a multi-agentic system design, involving more than one LLM call. The market's focus on agents is great because it's exactly what we've been talking about, even if we didn't have the perfect name for it initially.

Overall, the journey has been a learning experience. Thankfully, I had startup experience before this. At the start, we were heavy on exploration and keeping up with the market, which meant a lot of research. We still maintain a big research component, but as we've grown, we've brought in seasoned leaders who add maturity to deploying our solutions to data teams effectively. It's about transitioning from a research-heavy approach to a more balanced one, keeping research running in the background due to the fast-moving space of the industry. Hiring an amazing team with the right experience has been crucial.

How do you anticipate Numbers Station and the BI space as a whole progressing in the next 6-12 months? Any key priorities that you’d like to highlight?

I think there will be huge consolidation in the BI and data analytics space. The modern data stack has seen an explosion of tools, but BI alone won't be compelling enough. We'll see a move beyond descriptive analytics to predictive and causality, as well as downstream actions. We aim to lead in this consolidation by accelerating existing tools and focusing on the bigger picture. Expanding into actions and workflows will be key for us over the next six to twelve months.

Lastly, tell us about the team. Are you hiring, and what do you look for in prospective team members?

We're definitely hiring and always looking for great candidates. The biggest thing we screen for is a growth mindset. It's important to know what you know, but it's just as important to be able to learn quickly and branch into new areas. This is crucial in the fast-moving AI space. We need people who are continually curious, willing to work across the stack, and excited to be part of a high-growth company.

Conclusion

To stay up to date on the latest with Numbers Station, follow them on X and learn more about them at Numbers Station.

Read our past few Deep Dives below:

• Zilliz Cloud is powering vector search for the enterprise 🔋

• Kolena aims to redefine quality standards in ML 🎛

• Pinecone Serverless is turbo-charging your time-to-prod ⚡️

• Cleanlab is automating data curation at scale 🌐

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

Join Slack | All Events | Jobs