Full of Intellegens: Meet the Michelin chefs of the AI world
The proliferation of Intellegens’ applications amounts to a litany of technology at the edge of science. Drug discovery, batteries, manufacturing, software and data, chemicals, materials – how does one company get to be working in so many sectors?
The company spun out of the University of Cambridge in 2017. It has created an AI-based toolset that trains deep neural networks from sparse or noisy data. The technique was initially developed at the Cavendish Laboratory and is encapsulated in Intellegens’ first commercial product, Alchemite.
Alchemite’s algorithms “can see correlations between all available parameters, both inputs and outputs, in fragmented, unstructured, corrupt or even noisy datasets”. Prediction of missing values, error-detecting and optimised target properties result in accurate models that “clear data analysis bottlenecks, reduce the amount of time and money spent on research, and support better, faster decision-making”. And all that can be based on data that is as little as 0.05 per cent complete. Incredible.
Ben Pellegrini and Dr Gareth Conduit co-founded Intellegens.
“We develop industrial formulations or recipes, like a Michelin chef,” Ben says. “A chef mixes things up – he knows how to make it special, and that process happens in materials, it could be alloys, ceramics, polymers, batteries, plastics, drugs... you treat them in a certain way and it will result in certain behaviours. That could be to make it taste sweeter, or be more absorbent, or lighter – we help guide them to get these results quicker.
“All have a concept of a recipe or a formulation, so for instance batteries have a maximum power and weight, or how quickly they charge, or the level of power they achieve.
“And currently that’s all being developed by academic or specialist R&D departments. We provide a specialist tool, not based on the experience of the person working there, to guide the process and make it more transparent.
“For instance, silicon added to steel makes it harder, so what else could you apply to enhance that process?
“The current team might expect the formulation to look a certain way – that has been one of the problems in the past, and where you suggest radical changes people are sceptical and nervous, and with good reason.”
Intellegens – named in Cofinitive’s emerging #21toWatch longlist this year – started well. Prospective clients were immediately interested.
“Our USP is this ability to apply deep learning to sparse and noisy data,” Ben says. “We initially spent 12 months working out where to apply what we’d developed, so for instance for Rolls-Royce, new alloys need all sorts of physical characteristics, and that might involve $20m for testing, and Gareth got that figure down to $2m.
“Deep learning had not been applied to that area. Before, people did different things in different ways. We applied a new method approach suitable for all sorts of sectors.
“We started as a consultancy for deep learning to solve a company’s particular problem – and that’s still going on – and on the back of that developed the software.
“The model was developed in-house and now we have our first product out and it allows models to be deployed and trained in certain circumstances. Big data sets can be quite expensive.”
The rapid progress has involved some steep learning for in-house too. There’s a team of 10 based at the company’s base at Barclays Eagle Labs in Chesterton Road, Cambridge. Finance from Cambridge Enterprise and Wellcome’s Pathfinder awards “and a local angel” have helped develop a functional revenue stream pretty early.
“At the moment we’re very lucky to be generating revenue with support to build the business on customer revenue, which is one of our key targets,” Ben says. “Investment is always an option but we’re doing OK. We might reconsider if a bigger project came along.”
Progress, as always, involves surfing the crest of a wave and in 2017 the government announced the Faraday Challenge as the transport system gears towards an electric future. The challenge, part of the Industrial Strategy Challenge Fund, funds the research and development of electric car batteries.
“The Faraday Challenge is a five to 10-year programme,” says Ben. “The key as we’re moving towards the electric transport era will be batteries. Even Boeing has worked out how to do that for an electric plane engine.”
Indeed Boeing’s Passenger Air Vehicle (PAV), designed to carry passengers, and a smaller Cargo Air Vehicle (CAV) are lifted by several battery-powered propellors. The goal, says Ben, is “to scale up the manufacturing process – some materials develop differently and at the moment it’s trial and error, we use sparse data to extract maximum value”.
Similar leaps in R&D potential are taking place in drug discovery as Alchemite evolves its market.
“Alchemite is up and running,” Ben concludes. “It’s not an off-the-shelf purchase, it might change in a couple of places, it might change in big-enterprise R&D departments. There’s an on-boarding process taking place, to understand the technology and to make sure we’re solving the problem and that the problem needs solving.
“A lot of deep learning solutions are falling over because of the quality of data so we sit quite nicely in the mix.”