The Challenge

Slow Feedback Loop and Expensive Data Science Notebook Infrastructure

Yonder had outgrown their practice of using Jupyter notebooks on individual data scientists’ computers shared via GitHub with embedded Snowflake credentials that even data engineers struggled to keep running. They tried standardizing notebooks in Docker containers as well as using JupyterHub to manage their notebooks, but these solutions couldn’t support the demands of their team.

“Our notebooks became unwieldy and hard to manage which dramatically slowed down our development cycles and created barriers to getting business value out of our data science efforts.”

– Taylor McCaslin, Senior Product Manager at Yonder

About Yonder

About Yonder Yonder is a fast-growing authentic internet company on a mission to give the online world the same amount of authentic cultural context as the offline world. Using artificial intelligence and machine learning, Yonder identifies the groups and narratives that drive online conversations, helping organizations determine which matter most and least and creating the confidence to act. Yonder has helped identify Russia’s campaign to influence the 2016 presidential election, uncovered inauthentic behavior on user-generated content sites like Rotten Tomatoes, and helped Fortune 500 brands contextualize and understand factions and narratives on public social media platforms like Twitter and Facebook.

Yonder has built their entire data pipeline and ETL processes on top of Snowflake, ingesting and analyzing ~20 million pieces of social content daily, with a current database of over 3 billion records. Their data science and analytics team consists of both technical and non- technical people who need to collaborate across the product development lifecycle to provide timely insights to customers about online activity.

Yonder Gets to Market 4X Faster with Zepl and Snowflake

4x faster time to market

$75k annual savings identified in first month

Out-of-the-box security controls to Snowflake

Yonder-4@2x

The Challenge

Slow Feedback Loop and Expensive Data Science Notebook Infrastructure

Yonder had outgrown their practice of using Jupyter notebooks on individual data scientists’ computers shared via GitHub with embedded Snowflake credentials that even data engineers struggled to keep running. They tried standardizing notebooks in Docker containers as well as using JupyterHub to manage their notebooks, but these solutions couldn’t support the demands of their team.

“Our notebooks became unwieldy and hard to manage which dramatically slowed down our development cycles and created barriers to getting business value out of our data science efforts.”

– Taylor McCaslin, Senior Product Manager at Yonder

The Solution

“Snowflake is at the heart of everything we do at Yonder. Every part of the company uses analytics powered by Snowflake.”

– Chip Young, Chief Data Architect at Yonder

Zepl works natively with Snowflake, allowing Yonder data scientists and analysts to interact with data in real time and prototype new analysis daily.

Yonder connects to Snowflake through Zepl’s data science platform, enabling easy sharing of notebooks, visual analytics and machine learning models while supporting Yonder’s ability to tell customers the story of their social data. The combined efficiency of Snowflake and Zepl empowers Yonder to create new features faster and provide customers with better contextualized and actionable insights.

“This has democratized access to our data science. By allowing real-time analysis and rapid prototyping on production Snowflake data, we’ve been able to get new features to market 4x faster than before.”

– Taylor McCaslin, Senior Product Manager at Yonder

Speed Up Research Cycle 4X

“Now our data scientists and analysts can run experimental algorithms on top of our production Snowflake data and can take the results to our Client Success Team. We’re able to get high fidelity feedback using actual client data to understand if our data science is working the way we expected. We’re able to discover client value faster, and quickly iterate on our algorithms, which has accelerated our time to market.” said Taylor.

“People think that open source tools like JupyterHub are free. In fact, they end up being very expensive… both in people hours to maintain, infrastructure costs, as well as lost productivity when they inevitably have downtime.”

Identified $75k in Annual Savings within the First Month

“Zepl solves all that infrastructure headache by providing a SaaS solution that manages all our notebooks in the cloud, with each notebook running in an isolated container,” said Taylor.

“Zepl allows us to focus on data science, rather than managing infrastructure.”

Similar to Snowflake pricing, Zepl’s pay-as-you-go pricing model automatically provisions compute resources for the workload size. Yonder no longer has to worry about over- or under-provisioning their local or network hardware, and they are only charged for the resources they use. Within the first month of using Zepl, Yonder identified $75k/year savings in maintenance overhead and downtime.

Out-of-the-box Security for Snowflake Credentials as an added security bonus: Yonder no longer needs to embed Snowflake credentials directly into share notebooks and rely on configuration file tricks to prevent credentials from being pushed to Github.
Zepl’s secure keychain allows Snowflake credentials to be managed outside of the notebook and controlled at the user level, providing out-of-the-box security for all their notebooks.