Deepset raises $30M to help enterprises unlock the value of LLMs

Germany-based Deepset, a startup that helps enterprises unlock the value of large language models (LLMs) in their workflows, today announced $30 million in a fresh round of funding. The company said it will use the capital to further develop its commercial offering, deepset Cloud, with new capabilities, including optimizations for virtual private cloud (VPC) setups and a focus on the observability side of things.

The investment has been led by Balderton Capital with participation from existing company investors GV (Google Ventures), Harpoon, System.One and Lunar. It takes the total capital raised by Deepset to $46 million. Previous backers of the company include Snorkel AI’s Alex Ratner, Deepmind’s Mustafa Suleyman, Cockroach Labs’ Spencer Kimball, Cloudera’s Jeff Hammerbacher and Neo4j’s Emil Eifrem.

The development comes at a time when companies across sectors and around the world are looking to leverage the power of large language models in their internal systems to better grapple with the challenge of growing data volumes and make their teams function more efficiently. 

How does Deepset provide LLM support?

By 2025, the global datasphere is expected to grow to 163 zettabytes. Of this, the data from enterprise systems will be nearly 60%, meaning teams will have a mountain of data to deal with. This will make searching, retrieving, summarizing and analyzing relevant, work-related information quite a task.

Now, while large language models can be trained to develop linguistic intuition and semantics, using them to address this challenge in business applications can be quite a task, especially for non-AI companies. This is where Deepset comes in.

The company offers an open-source framework — dubbed Haystack — that allows developers to choose components required for a modern NLP project, from proprietary and open-source LLMs and vector databases to file converters and text embedding models. Once the components are picked, the framework plugs them into pipelines or agents to build LLM-driven applications. 

Such an application could be anything from a Google-like search engine for company documents to conversational AI to a powerful internal helpdesk.

Deepset started off five years ago with bespoke NLP solutions and followed those efforts with the launch of Haystack in 2019. Last year, it expanded its portfolio with deepset Cloud, a model-agnostic, SOC 2-certified cloud platform that lets AI teams build customized and flexible LLM systems while retaining full ownership of their data. This is the company’s commercial offering.

As the company explains, deepset Cloud covers the entire life cycle of a modern NLP application — experimentation, production and observability — while making it easy to compare and exchange different language models. It gives stakeholders a unified environment to work with fast prototyping, frequent feedback cycles and easy customization.

“Enterprises can see huge benefits from leveraging LLM technology. At Deepset, we’re providing a platform that helps to bridge the decades of research in machine learning and computer science into production-ready applications. 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,” Milos Rusic, cofounder of Deepset, said in a statement.

The 50-strong company works with enterprises across the U.K., Europe and the U.S. Legal publishing house Manz, for example, was able to use deepset Cloud for developing LLM-enabled products aimed at helping find precedents, relevant regulations, templates and more from millions of documents. 

Meanwhile, aircraft maker Airbus’s R&D team is using Haystack to build an application that helps pilots discover and use the most relevant aircraft operation guidelines right from the cockpit. The open-source framework has seen a 250% increase in active users, according to the company.

Plan to do more

With this round of funding, the company plans to continue international expansion and develop its products with new capabilities.

“We’ll do this by refining deepset Cloud for RAG (retrieval-augmented generation) applications: in particular, improving the evaluation of every component in a RAG pipeline. We’ll also focus on making the platform viable for customers with heavy privacy constraints by optimizing for virtual private cloud (VPC) setups,” the company said in a blog post.

In addition, it will focus on diversifying and improving LLM observability in deepset Cloud, giving customers confidence in the performance of their LLM applications in product environments, the blog post said. 

Source