Welcome to Snowflake’s Startup Spotlight, where we learn about awesome companies building businesses on Snowflake. In this edition, find out how Hum is applying the power of AI and large language models (LLMs) to help publishers build stronger customer relationships, and how the mantra of “build what people want” helped their leadership team make the decision to target publishers as their target audience.

Tell us about yourself, Dustin. What inspires you as a founder?

I’m a Michigander turned Virginian, curious technologist, former skunkworks innovation consultant and AI optimist. Traveling over hard ground on the way to building something important is what inspires me.

Explain Hum in one sentence.

Hum is harnessing frontier AI to transform content and audience data into actionable insights and personalized experiences.  

What problem does Hum aim to solve, and how are you using data to address the issue?

Publishers need to build direct relationships with everyone in their audience, not just pump out content. To do that, they need rich data and powerful AI. Hum provides both from a team that mixes experts in publishing, big data, AI/ML, marketing and UX.

Hum transforms all content, people, topics and behaviors into embeddings, the native language of large language models, so that the LLMs at the core of Alchemist (Hum’s AI engine) can power features such as infinite topic affinities, audience deep search, and next-generation personalized content recommendations.

How does Snowflake help you push the envelope in your line of business?

Hum’s fast data store is built on Elasticsearch. Snowflake’s relational database, especially when paired with Snowpark, enables much quicker use of data for ML model training and testing. We also use Sigma, paired with Snowflake, to power over half of the Hum dashboard experiences. We can spin up a Sigma workbook and embed it into the interface so users are able to see data insights clearly. 

When you were implementing Snowflake, how did you decide which architecture to use? 

Because we collect and manage our customer’s data, we have a managed architecture. While some of the data we collect comes from existing systems such as a CRM or an EMS, first-party data that’s being collected from websites only lives in Hum. However, we still democratize this data in two ways: Embedded Analytics with Sigma Computing and Snowflake Secure Data Sharing. The embedded analytics provide our users with a simple interface to investigate and learn from their data. Those who want to dig deeper can use the Secure Data Share to do their own analysis and data science without going through a cumbersome extraction process. 

What do your customers think about this model? 

This approach has been really well received by our users since it frees people to work in the tooling that makes the most sense to them.

Snowflake Secure Data Sharing helps reinforce the fact that our customers’ data is their data. While most customers prefer the Hum dashboard or APIs, more advanced customers want to flow more of the raw data into their warehouses or lakehouses. Snowflake makes it easy and cheap for them to pull in their data.

Tell us more about how you’re using Snowflake and Sigma for your dashboards.

Our fast data store is a document database, so we lean on Snowflake for analytics workloads. By pairing Snowflake with Sigma Computing for embedded analytics, we can ship agile dashboards fast. We use this to iterate on new product features and deliver custom views for more variable use cases. It’s amazing to be on a customer call, make a product adjustment in the background, refresh the screen, and have the update appear immediately. We can only work at that speed thanks to Sigma and Snowflake.

What’s the most valuable piece of advice you got about how to run a startup? 

Build something people want. It’s the YC mantra, but it definitely helps cut through the noise of growth hacking, VC pressure and fads.

What’s a lesson you learned the hard way? 

As I said previously, build something people want and pivot more quickly. We originally targeted the professional association market, and they’re just not ready for Hum. Pivoting to publishers and other content-rich organizers meant we were selling to people who “got” what we were doing, were very interested, and made good customers. They were fully using the product, generating value from Hum, and also giving us good feedback on future improvements. 

Given the current boom in AI applications, what role will Hum play in the industry’s future? 

Hum intends to be the brain and nervous system for publishers, giving them the ability to know, interact with, and build relationships with their end users. In 3-5 years, this will be essential infrastructure. Without it, publishers will fall prey to big aggregators like ResearchGate and the big tech giants (who will charge them for access to their own significant audiences). With it, they have the chance to “own” their own audiences and remain independent and thriving for the foreseeable future.

The post Startup Spotlight: Hum Applies AI and LLMs to Help Publishers ‘Own’ Their Audiences appeared first on Snowflake.