Stack AI wants to make it easier to build AI-powered workflows

Stack AI co-founders Antoni Rosinol and Bernardo Aceituno were PhD students at MIT completing their degrees in 2022, at a time when large language models were becoming more mainstream. ChatGPT would be released to the world at the end of the year, but even before that, they recognized a problem within companies combining data with models without much expertise and knowledge – and they wanted to change that.

After graduation, they moved to San Francisco and joined the Winter 23 team at Y Combinator, where they started Pile and improved their idea. Today, the company has created a low-code workflow automation tool designed to help companies build AI-driven workflows, including, for example, chatbots and AI assistants. The company has raised $3 million so far.

“Our platform allows people to create workflows that require connecting different tools to work together. We focus on connecting data sources and LLM, as this way you can create powerful workflow automations. We also offer many other tools and features to automate complex business processes,” Aceituno told TechCrunch. They’ve only had a working product for six months, but they already report over 200 customers using the product.

Essentially, this involves moving elements onto a workflow canvas. This typically includes a data source such as Google Drive and an LLM along with other workflow components such as a trigger or action component to build the workflow, allowing the customer to build AI programs without a lot of coding. The coding itself is not AI-based, but the tasks in the workflow often are and could require some manual coding to make the workflow run smoothly.

Some of their first customers are in the healthcare industry, and Aceituno acknowledges that they need to be careful with applications that involve doctors and patients, especially when internal data sources are not always reliable or may contain contradictory or outdated information.

In these cases, he says, it’s important to rely on the human expert, the doctor, to make the call on the quality of the response. As another layer of protection, they include source citations in each answer so that the health professional can check the source before accepting the answer.

“That being said, it’s true that you can put in rubbish and then the reports will also be rubbish and that’s why these assistants are required not to take over the whole process,” he said.

Coming straight from MIT and launching a startup, Rosinol says that going to YC really helped them understand the business side of things and how to improve their startup idea by working with clients.

“We started with an initial version of this API that was much more developer-focused. And we started with a few customers with the idea that we wanted to use AI to automate RFP responses or automate sales. And working with clients, it became very apparent that the real challenge was not training a model, but effectively searching and connecting data sources to those language models.”

The company currently has six employees, but is hiring engineers and sales and marketing professionals.

The $3 million investment closed about a year ago. Investors include Gradient Ventures, Beat Ventures and True Capital along with LambdaLabs, Y Combinator, Soma Capital and Epakon Capital.