TL;DR

Ferret is an AI platform architected from the ground up to empower companies and individuals with real-time, unbiased intelligence to identify risks and embrace opportunities. Ferret partnered with Anyscale to scalably process and integrate thousands of datasets while simplifying management and iterating more quickly.

Overview

Ferret is an AI platform architected from the ground up to empower companies and individuals with real-time, unbiased intelligence to identify risks and embrace opportunities. 

Ferret is using AI to disrupt and transform an information market traditionally dominated by companies like LexisNexis and Experian. Leveraging cutting-edge AI and hundreds of thousands of global data sources, Ferret provides information — once only available to the financial industry — to all types of professionals, from angel investors to business leaders, making transparency the new norm. For a fraction of the cost of traditional approaches, Ferret can offer cutting-edge relationship intelligence using publicly-available sources. This allows Ferret’s customers to establish more effective relationships.

The Challenge

Ferret’s defining innovation is to integrate thousands of data sources to inform individuals and businesses with relationship intelligence.

This involves a technique often called “identity disambiguation” - trying to identify the same user across multiple datasets without a common unique identifier. For example, “Tim Smith” is a very common name across many social media platforms, so you can’t just pull together every post by “Tim Smith” to understand his social activity. And Ferret has to do this across thousands of sources, often pulling updates every hour.

The Ferret team originally assembled a foundation using NoSQL database MongoDB as well as AWS OpenSearch, Apache Kafka for a listening service, and spaCy for Natural Language Processing. The system worked, but the team struggled with the infrastructure complexity, and faced challenges and delays when attempting new experiments.

The Solution

Anyscale simplifies and streamlines development and operations while providing a future-proofed platform for ongoing AI innovation.

Anyscale provided a unified framework to scale AI workloads without requiring the team to maintain a custom stack of open source components to manage their ever-growing array of data sources. ”My job is to simplify deployment complexity,” said Al Basseri, CTO at Ferret. “Having Data Scientists doing MLOps work doesn’t scale, so we took that off their hands."

Anyscale’s platform lets the Ferret team iterate more quickly, trying new models, datasets, and even use cases much more quickly than before.

Ferret regularly processes billions of records, sometimes as frequently as hourly. Anyscale distributes processing across Ferret’s CPUs and GPUs, maximizing utilization by making sure, for example, that training jobs running on a GPU aren’t constrained by slow data ingestion.

“With the efficiency of the Anyscale Platform, I can scale my users and business much more quickly than my cloud costs,” said Al Basseri.

The Impact

Using Anyscale, Ferret simplified its infrastructure, gaining agility, reducing costs, and ensuring maximum flexibility for the future.

Lower Costs: Anyscale’s efficiency reduced Ferret’s ongoing cloud costs by 20%

Faster Iteration: The Ferret Data Science team now iterates 3-4X faster thanks to simplified infrastructure and “fewer moving parts.”

Effortless Scale: With an AI Application Platform in place, Ferret can scale to add new users, models, and datasets without adding infrastructure.

Future Proofing: Given how quickly AI technology is evolving, the team at Ferret didn’t want to make any choices that would limit future flexibility or leave them locked-in to a single cloud platform or toolset. “You don’t want to spend time creating a half-inch drill and wake up to a world that’s gone metric,” said Al Basseri.

Moving forward, the team at Ferret is looking into Large Language Models and is currently prototyping, including fine-tuning. In Al's words, “What’s great about having a platform is that we can add new AI use cases like LLMs without deploying and integrating more infrastructure.”