HerculesAI has been working with large languge models long before it was cool

HerculesAI has been working with large languge models long before it was cool

HerculesAI (formerly Zero Systems) has been working at automating professional services since 2017, originally concentrating on the legal industry. As part of that, it has actually been building large language models for several years, long before the idea entered the public consciousness. As such, it found itself in the right place at the right time when ChatGPT popped onto the scene in late 2022, and suddenly everyone was talking about LLMs.

Today, the company announced a $26 million Series B investment to help keep building on its recent momentum.

Alex Babin, company CEO and co-founder, says that they had been working on small models since around 2020 with half a billion parameters to 2 billion parameters and running them on edge devices for compliance purposes, but prior to the emergence of ChatGPT nobody paid much attention to that aspect of their solution.

“It was maybe eight or nine months before ChatGPT, and I remember speaking to our clients, explaining to CIOs what an LLM is – and no one cared,” Babin told TechCrunch. By November that year, of course that would rapidly change and suddenly everyone was interested in the concept. That shift has helped drive rapid growth in the business over the last year.

Today, the company has several models performing three key functions: intelligent data extraction, data transformation and data verification. The first is pretty standard and involves pulling data from documents. The second part builds a set of rules and structures around that data automatically, but the third part, verification, is particularly important, he says.

“It’s really the Holy Grail when you can compare information extracted and then transform it to the source of truth, whether that’s regulations, policies, contracts, laws or anything,” Babin said. That ensures that any issues that conflict with the source materials are flagged for employees automatically.

Those three buckets have also enabled the startup to build a multi-agent system on top of those services to help automate all of these activities. “Those multi-agent systems can be applied to high value, continuous processes or workflows that require [automated] decision making,” he said.

For his core regulated industry customers all of this is particularly important. Today, that’s not only legal, but also insurance and financial services.

Their AI strategy appears to be working with the company reporting 4x growth over the last year. They count 30% of the top 100 law firms in the U.S. as customers. They also have a slew of other Fortune 500 customers including Mercer, Standard & Poor’s and State Farm.

The company currently has around 75 employees, but in spite of the additional money, Babin says he is planning to stay lean and invest more in refining internal processes than adding employees. “I don’t see why we need to hire more people. We will actually invest more in our internal processes and automation. We have to eat our own dog food and use our own products to make ourselves more scalable,” he said.

Today’s funding was led by Streamlined Ventures with participation from Proof VC, Thomson Reuters Ventures, Alumni Ventures and various industry angels.