Activeloop, a member of the Y Combinator summer 2018 cohort, is building a database specifically designed for media-focused artificial intelligence applications. Today the company announced a $5 million seed investment led by 468 Capital and CM Ventures, with participation from Tribe Capital, Shasta Ventures and various tech industry angels.
Davit Buniatyan, founder and CEO at the company, says the company developed out of research he was doing at Princeton where he saw the need for a streaming database of unstructured data like images and video specifically designed for AI use cases. The company is also launching an alpha version of a commercial product today.
"[We are] introducing a database for AI, specifically a storage layer that helps to very efficiently store the data and then stream this to machine learning applications or training models to do computer vision, audio processing, NLP (natural language processing) and so on," Buniatyan explained.
In practical terms, this takes the form of an open source API that enables users to convert the data (video, audio, etc.) into a mathematical representation that machines can understand. In addition, the API lets you track different versions of the data and finally lets you store that in a repository like Amazon S3.
By streaming the data, the data scientist or developer no longer has to download all of this data to a local laptop, but sends it through for use by the model in the same way that Netflix streams a movie to you, so you aren't downloading the whole thing to your local machine.
Activeloop image database. Image Credits: Activeloop
He says that there are 55 contributors to the open source project and 700 community members overall. The program has been downloaded 300,000 times so far.
He came up with the idea for streaming the data while working on large amounts of data at the neuroscience lab at Princeton. He found the files were so huge, it wasn't practical to download them, so he came up with this way of streaming them that would become the basis for Activeloop.
The company has 15 employees so far and is looking to hire around half a dozen high-end engineers. He says as he builds the company he is looking at diverse talent wherever it may be. He points out that his VP of engineering is a woman. Recently the company offered a job to an Ethiopian engineer who ended up turning it down, but he says it's an example of looking for talent wherever it happens to be.
"It's not about where a person is originally from, or what kind of biases in terms of where in the world he or she was born. What matters is that we are on this mission. We are building this [company] to have an impact, and we do care about people on our team who are building this," he said.
The API is open source, but the company's commercial product, which is being released in alpha today, will be adding features on top of this, such as being able to run SQL queries on the streaming data, which aren't available in the open source product. The company is pre-revenue at the moment, but is hoping to release the commercial version in Q1 next year.