This code optimisation startup can reduce AI’s carbon footprint by 50%
As the global artificial intelligence market continues to grow exponentially, with a record 65 companies in the space reaching unicorn status last year and more organisations than ever incorporating AI tools into their everyday operations, so does the need to develop more efficient AI solutions.
“Take any smartwatch, for example,” Leslie Kanthan, CEO and co-founder of code optimisation startup TurinTech, says. “Its battery lasts however long the code is running — say about 20 hours. But if you have a way to improve that code so that it’s consuming less power, then the battery on your watch will last longer.”
Launched in 2018 as part of Conception X Cohort 2, TurinTech has found a way to address inefficiencies in code feeding AI models to get them built rapidly while reducing compute time and making them as energy efficient as they could be.
Researchers have previously suggested that training a single AI generates roughly five times the lifetime emissions of an average car. The startup’s AI platform, evoML, can reduce carbon emissions by 50% by optimizing models for quicker inference speed.
“Given any piece of code, we can reduce memory and energy consumption whilst maintaining performance,” Kanthan says. “The applications there are tenfold.”
From helping data scientists build models that use MRI data to predict diseases in days rather than weeks or months, enabling teams to focus on improving the model’s accuracy from the start, to pushing profitability up for hedge funds through code optimisation of deep learning models moving large sums of money, TurinTech’s approach has already proven successful time and again.
One year after launch, the startup raised more than £1 million in a pre-seed funding round led by IQ Capital, and closed a £3 million stealth round in 2021 with Speedinvest’s support. The team has also received two offers of acquisition to date, and is about to close a large Series A funding round.
“We see the journey moving us into a billion-pound company,” Kanthan says. “We see our growth trajectory increasing because the AI market is growing, products are evolving quickly, users are becoming more AI literate, and there’s a lot more code in this space than there ever has been. Our journey is correlated directly with that.”
Back when it all started, TurinTech was a small team of four researchers working out of a desk at UCL to solve a problem they had all witnessed on a daily basis: in the data science space in particular, code was getting bigger and bigger and data scientists were often writing new code over old code, losing track of how to make it more efficient, nimble and well structured.
“There was very little knowledge back then about where you could go to get help with your idea,” Kanthan says. “A few professors knew we were working on a product, told us about this incubator that was also spinning out of UCL around the same time, and encouraged us to apply.” That’s how the team joined Conception X.
Here, the team became familiar with the VC space and met other PhD startups that were further along on their growth journey, learning from their experiences. “The mentorship itself was critical,” Kanthan says. “I think that we would be here without Conception X, because a startup’s journey is also based on conviction and the team’s willpower, but the journey would have been much more difficult. We would have had a lot of problems — teething issues — that we learnt to avoid because we had gone through the programme.”
Pitching to investors was one of the most valuable early lessons the team learnt on the programme. “If investors don’t understand what you’re saying within a few minutes, even if you have a fantastic idea, they’re likely to pass,” Kanthan says. “Before Conception X, we weren’t funded; after Conception X, we were funded. One can draw their own conclusions from that.”
Fast-forward four years, and TurinTech can now rely on a team of 50 people working with clients from different industries, including banks, hedge funds, database companies, supermarkets and more. One example is TurinTech’s partnership with database company Exasol to advance their analytics by building high-performing models that run four times faster and deliver more accurate insights.
Asked if he has any advice for PhD founders who are just starting out, Kanthan says: “Get your research idea into a coherent business plan. Get the best talent to work with you, attract people through good partnerships and relationships. Give back as well, as that supports the very structures that supported you. And get your first closed deal — a three-year license deal, for example, which is typically what we get from customers. That’s the validation, that’s what shows success. There are a lot of startups that continue in the space for quite some time without making money, and the journey becomes that much harder.”