TrueFoundry Inc., a startup with a platform for managing artificial intelligence workloads, today announced that it has raised $19 million in funding. 

Intel Capital led the Series A round. Eniac Ventures, Surge and Jump Capital chipped in as well. TrueFoundry says that its institutional backers were joined by several angel investors, including former Cohesity Inc. Chief Executive Officer Mohit Aron.

Developing the software scaffolding necessary to run AI models in production can take months of work. TrueFoundry’s namesake platform promises to ease the task. The software includes a set of prepackaged features for deploying and maintaining large language models, which spares developers the hassle of building everything from scratch. 

Enterprises that build their own AI models often maintain cloud-based VS Code and Jupyter Notebook environments. The former tool is a popular code editor, while the latter eases the development of neural networks. TrueFoundry automatically shuts down VS Code and Jupyter Notebook environments after they’re no longer needed to cut infrastructure expenses. 

The company says its platform can also help enterprises manage their LLMs.

According to TrueFoundry, its software enables developers to turn a newly created LLM into an application programming interface with one click. The API makes it easier to integrate into the model with other workloads. Under the hood, TrueFoundry automatically adjusts the amount of infrastructure available to the API based on workload demand.

The platform can reduce LLM costs by making use of spot instances. Those are virtual machines that run on unused hardware in cloud providers’ data centers. Spot instances cost significantly less than standard virtual machines, but have technical limitations that can make using them difficult. 

TrueFoundry includes a library of more than 30 open-source LLMs. A fine-tuning tool allows customers to optimize those LLMs for their requirements by supplying them with custom training data. TrueFoundry uses a fine-tuning technique called QLoRA that trades off some training speed for a significant reduction in memory usage.

After customers deploy an LLM, the platform can monitor it for technical issues. TrueFoundry tracks metrics such as inference cost, latency and error rates. 

“Enterprises using TrueFoundry have built and launched their internal AI platforms in as little as two months, achieving ROI within four months – a stark contrast to the industry average of 14 months,” said TrueFoundry CEO Nikunj Bajaj (pictured, right, with co-founders Abhishek Choudhary and Anuraag Gutgutia.) 

The company also disclosed that its platform has been adopted by more than 30 customers. The list includes Nvidia Corp., which is using the software to optimize its LLM workloads’ graphics card usage. TrueFoundry claims that customers have deployed more than 1,000 AI clusters powered by its technology over the past year. 

The company will use its Series A round to build new features and accelerate go-to-market efforts.

Photo: TrueFoundry