Eventual launches with $30M in funding to streamline multimodal data processing – News

Multimodal data processing startup Eventual Inc. is looking to transform the way companies deal with unstructured data after closing on $30 million in venture capital funding.

The startup said today it has closed on the $20 million Series A funding round led by Astasia Myers from Felicis, with strategic participation from M12 Ventures and Citi. It also announced a previously undisclosed $10 million seed funding round, led by Brittany Walker from CRV, Y Combinator, Essence VC and Array Ventures.

Eventual sees itself eventually becoming the infrastructure foundation for multimodal data processing. It’s the creator of an open-source processing engine that’s purpose-built to deal with the “messiness” of unstructured information, capable of dealing with hundreds of petabytes of images, videos, audio files, handwritten notes and social media comments, among other data types.

That processing engine is known as Daft, and it’s already being used in production by companies such as Amazon Web Services Inc., Essential AI Labs Inc. and Together Computer Inc., where it helps them to process petabytes of information for workloads such as AI models, autonomous vehicles and product recommendation systems.

Eventual founders Sammy Sidhu and Jay Chia hit upon the idea for Daft during their time working on autonomous cars at the ride-sharing company Lyft Inc., where they found themselves struggling to deal with mountains of multimodal data that couldn’t be processed efficiently using legacy tools.

In an interview with News, Sidhu explained that self-driving cars need to process enormous volumes of messy data such as 3D scans, video feeds, images, text and audio. But when they were doing this at Lyft, they realized that no tool existed that could process and understand all of those different data types at the same time.

As a result, Sidhu and Chia were forced to piece together various open-source tools to get to grips with it all. But this took time, and the systems they devised were extremely unreliable.

“Traditional engines fail because modern AI workloads are different,” Sidhu wrote in a blog post. “They run custom models, hit external APIs, and process wildly diverse data types. A 0.1% failure rate that’s acceptable in testing becomes catastrophic when processing millions of files in production.”

Now, Sidhu says, there are many AI developers who are facing exactly the same kinds of challenges. Every company that’s building AI applications needs to process enormous amounts of multimodal data, using tools designed for structured data such as web clicks and financial transactions.

Sidhu and Chia’s solution to this is Daft, which they helped to design while working together at Lyft. Daft is a Python-native open-source data processing engine that can work quickly across any kind of modality. When they set out to build it, the co-founders wanted to make Daft as impactful for unstructured data as the Structured Query Language was for tabular datasets.

According to Sidhu, he never thought to start a company at first, but when he was applying for new jobs, he was asked by several interviewers if they could build a similar system for their companies. And that’s what inspired him instead to reate Eventual, which aims to commercialize the Daft platform.

Alongside today’s funding round, Eventual has opened the waitlist for access to the Eventual Cloud platform, which is said to be a production-grade version of Daft that’s specifically designed for multimodal AI applications.

Sidhu said Eventual has the ability to process petabytes of unstructured data as if it was merely processing megabytes of information, taking in millions of documents, images and videos without any infrastructure hiccups. It allows AI teams to ship new features in days rather than months, since they no longer have to spend 80% of their time worrying about data processing.

With today’s funding, Eventual is now setting out to expand its engineering team, eyeing talent with expertise in distributed systems, product engineers who want to help define how to work with data in the age of AI, as well as developer advocates who can show the world how to process multimodal data effectively.

Images: Eventual