Startup Typedef gets $5.5M seed funding to build customized data pipelines for AI model workloads

Data processing startup Typedef Inc. wants to help companies accelerate the deployment of artificial intelligence application pilots after closing on $5.5 million in seed funding.

Today’s round was led by Pear VC, a specialist early-stage investor, with participation from Verissimo Ventures, Monochrome Ventures, Tokyo Black and several unnamed angels.

The startup is targeting what it says is a $200 billion market for AI infrastructure with a new data platform purpose-built for modern generative AI workloads. As far as Typedef is concerned, AI inference is set to become the most vital aspect of enterprise computing infrastructure in the coming years. That’s why it has set out to build a data platform that’s able to handle every kind of AI workload with the efficiency, performance and predictability demanded by serious organizations.

It describes its platform as a “modern-day AI data infrastructure” that’s designed specifically for the kinds of highly scalable data pipelines needed for large language models to perform semantic analysis. Aimed at application developers, it automates the management of the most complex properties of mixed AI workloads, including things like token limits, context windows and data chunking through a simple user interface with familiar application programming interfaces and relational model.

The main purpose of Typedef is to accelerate AI deployments with more rapid and iterative prompt and pipeline experimentation, so companies can more easily determine which workloads are going to provide value, and then get them deployed in production.

It’s aiming to solve a problem known as “pilot paralysis,” which describes the alarmingly high failure rate of AI pilot projects. Typdef cites a 2025 study of chief data officers by Informatica LLC, which revealed that a staggering 97% of companies experience difficulties in demonstrating the value of generative AI applications. The same study found that 67% of CDOs have failed to move even half of their generative AI pilot projects into production.

Typedef co-founder Yoni Michael (pictured, right, with co-founder Kostas Pardalis) said it has become extremely difficult for organizations to move their AI workloads into production in a way that’s predictable, deterministic and operational. The vast majority of such projects just linger in the prototype phase, he said, and one of the main reasons for that is the lack of a suitable data infrastructure.

“Legacy data platforms weren’t built to handle LLMs, inference or unstructured data,” Michael explained. “As a result the workaround has been a patchwork of systems, ageing technologies and tooling or DIY frameworks and data pipelines that are brittle and unreliable.”

That’s the problem Typedef is trying to remedy with its data infrastructure, which is tailor-made to handle the intricacies of LLMs and the unstructured data they feed on.

“[It is] a solution built from the ground up with features to build, deploy and scale production-ready AI workflows – deterministic workloads on top of nondeterministic LLMs,” Michael promised.

Typedef stands out for its simplicity too, with its entire platform being serverless, bypassing the need for customers to provision and configure their infrastructure. Instead, they can just download its open-source client library, connect their data sources and start putting together their data pipelines within a few minutes.

The proof is in the pudding, and Typedef said early adopters of its platform have a strong appetite for what it’s doing for them. Its customers include the insurance technology firm Matic, which works with more than 70 carriers in the U.S. The startup says Matic is using its platform to quickly create production AI workflows built on top of its customer support transcripts and policy documents, helping insurers to drive significant cost savings and improve their compliance posture.

Matic Chief Product Officer Lee Maliniak says he has become a very happy man since he started working with Typedef. “Typedef lets us build and deploy semantic extraction pipelines across thousands of policies and transcripts in days not months,” he explained. “We’ve dramatically reduced the time it takes to eliminate errors caused by human analysis, significantly cut costs, and lowered our errors and omissions risk.”

Pear VC Partner Arash Afrakhten said we’re quickly entering a new era in AI infrastructure, where model training is giving way to inference. He said companies need an easier way to create reliable, scalable and cost-effective LLM workloads with minimal complexity. “I’ve backed this team because they’ve lived the problem, know what’s needed, and have the added experience of running multiple data infrastructure startups to successful exits,” he said.

Image: Typedef