Artificial intelligence tooling startup Zipline AI Inc. today announced that it has raised $7 million in seed funding.

Early-stage investment fund Wing VC led the round. It was joined by Stripe Inc., Box Group and Exceptional Capital.

Developers prepare AI training datasets using software workflows called data pipelines. Those workflows collect records from different systems, combine them into a single dataset and fix quality issues. Once training is complete, data pipelines can be used to continuously refresh an AI model’s knowledge pool with new information.

Building AI pipelines manually can take months of work in some cases. According to Zipline AI, its namesake platform reduces development times to a few days. 

The platform is based on Chronon, an open-source data management engine developed by Zipline AI co-founders Varant Zanoyan and Nikhil Simha Raprolu. They built the tool while working at Airbnb Inc., which uses it to power its core AI infrastructure. The software has also been adopted by OpenAI, Uber Technologies Inc. and other major tech firms.

Data pipelines created with Zipline AI can collect information from a long list of sources. Those sources include data warehouses, databases that power production applications and real-time data streams. Zipline AI can also collect contextual information that an AI model’s users include in prompts.

After aggregating records from a system, the platform turns them into features. A feature is a condensed version of a dataset that is easier for AI models to process than the original. For example, developers might turn a set of temperature measurements taken over the course of a week into a single average number.

Zipline AI can organize the features that it creates into training datasets or make them available to AI models during inference. It also provides support for so-called temporally accurate backfills. According to Zipline AI, the feature can be used to keep training datasets fresh by regularly adding new information.

Developers can interact with the platform via a Python-based application programming interface. The API makes it possible to provide a high-level definition of a feature and have Zipline AI automatically generate the data pipeline needed to create it. Additionally, software teams can create copies of an existing production pipeline and use them to test changes.

“Building AI systems is still a long and complicated process for most companies,” said Zanoyan. “We’re tackling that complexity head-on, making it faster and easier for teams to develop and deploy AI products while reducing costs and improving observability and governance.”

Zipline AI will use its newly closed funding round to grow its team and finance go-to-market initiatives. 

Image: Unsplash