Deepset GmbH today announced that it has raised $30 million to enhance its open-source Haystack framework, which helps developers build natural language processing applications.

The Berlin-based startup will also use the capital to extend its commercial offering, a cloud platform called Deepset Cloud. The offering is partly based on Haystack. 

Balderton Capital led Deepstack’s latest funding round. It was joined by GV, Alphabet Inc.’s startup investment arm, and Harpoon Ventures. The raise brings Deepset’s total outside funding to $46 million.

Language models often can’t be deployed in production on their own. To become usable, they must be integrated with supporting components such as databases. Deepset’s open-source Haystack framework makes it easier for software teams to deploy such supporting components and thereby speeds up development projects.

In the enterprise, many artificial intelligence initiatives focus on using a language model to analyze the contents of a database. A company might, for example, build an AI that draws on knowledge from a product database to answer customer questions. Deepset says Haystack reduces the amount of work involved in connecting data stores to language models.

Neural networks sometimes struggle to process business data in its original form. As a result, companies must reorganize their data into a format that is simpler to analyze. Haystack promises to ease the task by providing tools that can change the structure of business documents before they’re processed by an AI model, remove any unnecessary information they may contain and perform other modifications.

The framework also promises to streamline several related tasks. According to Deepset, Haystack includes a feature that can automatically generate training datasets for machine learning projects. It also offers pre-packaged language models that are capable of answering user questions about a company’s business information.

The company generates revenue by selling a commercial cloud platform called Deepset Cloud. It combines the core features of Haystack with a set of additional tools designed to streamline AI application projects.

Many of Deepset Cloud’s features focus on making it simpler to run language models in production. According to the company, the platform allows developers to deploy a newly developed neural network to their company’s production environment in one click. From there, Deepset Cloud can automatically add or remove infrastructure resources as the neural network’s requirements change and monitor it for technical issues.

“We’re providing a platform that helps to bridge the decades of research in machine learning and computer science into production-ready applications,” said Deepset co-founder Milos Rusic. “In the same way you don’t need to know much about microchip architecture to write software, you don’t need to be an NLP or LLM scientific researcher to use our Haystack framework and Deepset Cloud.”

Besides enhancing Haystack and Deepset Cloud, the startup will use its new $30 million funding round to accelerate go-to-market initiatives. It also plans to expand internationally. The company reportedly expects to hire between 20 and 25 new employees by year’s end as part of the effort. 

Image: Unsplash