After studying pure maths at Trinity, machine learning expert Paul Hayes decided that it was time to move to London, go ‘deeper’ into AI and start his own company.
“We’re supplying picks and shovels in a gold rush.”
This is how machine learning researcher turned entrepreneur Paul Hayes describes Humanloop, a software start-up he co-founded that is helping enterprise teams build generative AI applications for their customers.
“AI is a new computing platform that requires a whole new set of tools and frameworks for builders to replace the traditional software development paradigms,” Hayes goes on. “Humanloop’s mission is to enable the safe and rapid implementation of AI across the economy.”
Based in London, Humanloop emerged in 2020 as a spin-out of University College London (UCL), where Hayes was doing a PhD, specialising in probabilistic deep learning.
Having done his undergraduate degree in pure mathematics from Trinity College Dublin, Hayes’s first dalliance with software engineering and the tech world at large started during a brief stint in San Francisco.
Digging deep into AI
Once he completed his degree, the now chief technology officer at Humanloop moved to London with two goals: go “really deep” into artificial intelligence and start a company at some point.
It was during his PhD at UCL that Hayes befriended co-founder Raza Habib, who is now the CEO.
“Our other co-founder and friend Jordan Burgess was working as a researcher at Amazon Alexa at the time and was also looking to start something new after his time at Entrepreneur First in London,” Hayes says.
“We were all particularly excited about the wave of research in natural language processing and wanted to build technologies that helped facilitate adoption in industry as the research continued to surge.”
Humanloop’s journey to becoming a successful start-up really began when it joined Y Combinator. Within three months, Hayes, Habib and Burgess had their first minimum viable product and paying customers. Soon after, the start-up got further backing from the likes of Index Ventures.
Today, product and engineering teams at companies such as Duolingo and Gusto use Humanloop software to collaborate on AI product development and evaluate performance to produce high-quality generative AI applications for their customers.
“A new software stack is emerging for enterprises who need confidence in deploying their AI solutions, and Humanloop is increasingly becoming core to this,” Hayes explains.
Essentially, the team has created a new type of development environment with foundation models as a “first class citizen” that allows even non-technical teams to collaborate on building AI features.
“Our technology has created a new framework for how to automatically evaluate and understand how your AI application is behaving both during development as your teams are experimenting, and once it has been deployed to production and real users are interacting with it,” he goes on.
“This then facilitates our systematic workflow for improving models using techniques like prompt engineering and fine-tuning.”
Keeping AI teams in the loop
The idea is to make the process easy for multidisciplinary teams of engineers, product managers and domain experts working at tech companies that are rapidly adopting generative AI as a core part of their business.
“Technical leaders at these companies are struggling to standardise how they build with generative AI and to gain confidence around deploying real use cases to production,” Hayes says.
“This will include questions and pain points around model choice, prompt management, versioning and deployment controls, evaluation and monitoring, integrating data sources securely and fine-tuning to improve performance.
“Humanloop provides the answers to these points with an integrated solution that also works natively with all the major open and closed source foundation models.”
So far, the start-up has kept its team small, with around 10 people. But things are beginning to pick up, with the creation of a second office in San Francisco and an active hiring campaign to scale up.
“We’ve recently made the shift to serving much larger and more mature tech enabled enterprises as they begin their AI adoption,” he says.
“This is where we are seeing a lot of demand as technical leaders are grappling with how they are going to adapt and are looking for key infrastructure to facilitate the deployment of foundation models across their organisation and products. And that’s where Humanloop comes in.”
Find out how emerging tech trends are transforming tomorrow with our new podcast, Future Human: The Series. Listen now on Spotify, on Apple or wherever you get your podcasts.
