Artificial intelligence inference startup Simplismart, officially known as Verute Technologies Pvt Ltd., said today it has closed on $7 million in funding to build out its infrastructure platform and help companies to deploy AI models more easily.

The Series A round was led by Accel and saw the participation of Shastra VC, Titan Capital, and high-profile angels such as Akshay Kothari, co-founder of Notion Inc.

Simplismart has created what it says is a “fast inference engine” that enables companies to optimize the performance of AI model deployments. The startup says it wants to be seen as a critical enabler of AI’s transition into mainstream enterprise operations. To do this, it’s looking to solve a number of challenges that prohibit enterprise adoption of AI, such as the performance tradeoffs many companies are forced to make.

In a blog post, Simplismart’s co-founder and Chief Executive Amritanshu Jain said enterprises increasingly want to adopt AI but struggle to realize much value out of it. Part of the problem is that it’s not easy for companies to deploy AI by themselves. One alternative is to use third-party application programming interfaces, he said, but they’re expensive and rigid and pose concerns about data security.

“Every company has different inference needs, and one size does not fit all,” Jain said. “APIs are not tailored to scale for bursty workloads and cannot tweak performance to suit needs. Businesses need to control their cost vs performance tradeoffs. This will be the primary reason for a shift toward open-source models, as companies prefer smaller niche models trained on relevant datasets over large generalist models to justify ROI.”

Jain argues that few enterprises want to “rent their AI,” but says many are forced to do so because owning AI is not easy. To deploy large language models in-house, companies are faced with significant hurdles around scaling their infrastructure, creating a continuous integration and continuous deployment pipeline, getting access to compute resources, model optimization and cost-efficiency.

At present, most companies use one of two off-the-shelf solutions for their AI, but these both have limitations. For instance, MLOps platforms enable orchestration and model serving, but they do not provide an optimized environment for AI in production, which means companies face severe performance limitations. The alternative is to use generative AI cloud platforms, or “GPU brokers,” which provide optimized APIs and performance, but they come with serious data privacy and cost concerns.

Simplismart’s inference engine is designed to give enterprises a new option, providing a standardized language that software engineers can use when creating generative AI applications. Its primary benefit is that it reduces the time it takes for models to respond to queries.

It cites benchmarks that demonstrate its ability to run the open-source Llama 3.1 8B model at a throughput of more than 440 tokens per second. This represents an impressive speed breakthrough, and it’s bundled with a comprehensive MLOps platform that’s tailored for on-premises AI deployments.

According to Jain, there’s a big market for what the startup is offering. He cites data that shows how almost 90% of enterprises’ machine learning projects never make it into production.

“The adoption of generative AI is far behind the rate of new developments,” the CEO said. “It’s because enterprises struggle with four bottlenecks: lack of standardized workflows, high costs leading to poor ROI, data privacy, and the need to control and customize the system to avoid downtime and limits from other services.”

Simplismart’s declarative language is similar to Terraform and helps software teams with tasks such as fine-tuning, deploying and monitoring generative AI models at scale. The platform helps to standardize all of these workflows, ensuring teams can optimize their models for performance.

Simplismart was founded in 2022 by Jain alongside Chief Technology Officer Devansh Ghatak. While Jain’s experience lies in cloud infrastructure, primarily from his time at Oracle Corp., Ghatak’s area of expertise is search algorithms, which was honed during his time at Google LLC.

In just two years, with less than $1 million in capital, Simplismart has managed to create a powerful MLOps platform for deploying models complete with a high-performance inference engine that the founders say is the world’s fastest. Companies can create, fine-tune, deploy and then run their AI models on-premises at suitably rapid speeds, boosting performance without the cost and security concerns.

Simplismart says it wants to help companies deploy custom generative AI applications with full control. It sees itself providing the granular Lego bricks companies need to create their own inference and deployment environments, so they can do that.

To date, Simplismart has amassed about 30 customers who are delivering a combined $1 million in revenue on an annual run rate basis. With the funding from today’s round, Jain thinks the company can reach $5 million by the first quarter of next year.

The money from today’s round will be a big help for Simplismart, and it’s earmarked for product development, recruitment and investment in its sales and marketing efforts.

Accel Partner Anand Daniel said more companies have begun to realize the merits of deploying and customizing AI models on their own infrastructure, such as control over performance, cost, data security, privacy and more.

“What blew us away was how their tiny team had already begun serving some of the fastest-growing generative AI companies in production,” he said. “It furthered our belief that Simplismart has a shot at winning in the massive but fiercely competitive global AI infrastructure market.”

Image: Simplismart