The Founder of Scholarcy: Phil Gooch
Phil Gooch is the Founder of Scholarcy which is helping more than half a million students and researchers read and understand academic texts more easily. I was lucky enough to sit down with Phil and discuss his experience working in the start-up industry and how that inspired him to create an EdTech start-up himself in 2018.
JMR: Can you tell me a little about yourself, and how you began working in the tech industry?
PG: When I was about ten years old my friend’s mum taught programming at the local sixth-form college, and they had this computer at home called a Commodore PET. My friend and I used to play games on it, and then his mum said she’d teach us how to program. Soon I was writing games, and then my brother got a ZX Spectrum, and I learnt to program that too. I wrote a few games and some programs to play music on it.
But when I became a teenager, I lost interest in computers completely.
Even at university, I chose to do an extra maths paper instead of the programming assignment! After university, I went to work in academic publishing, and we had 1 computer for the whole office, so I thought I’d better start making the most of it.
My job was to copy-edit academic manuscripts before journal publication, so I thought it might be cool to try to automate some of this. Soon I had developed a suite of tools that edited the manuscript and typeset it in Word with one click. I guess I was always interested in what was possible with text. I ended up working for several publishers helping them automate editorial and production processes.
In 2009 I went back to university to do a PhD and it was different to when I did my undergraduate degree back in the late 1980s. I had to read so many academic articles; they were very dry, boring to read, and just in this unattractive format. Also, the problem wasn’t finding them, it was organising and reading them. I figured it would be great if there was a better way of getting the knowledge off the page and into my brain easier.
After doing the PhD I went to work for various start-ups, and this was great as I quickly learned about the proper software engineering process, from building a model, organising into code that had some sort of interface, writing tests so you could prove it was working, and then deploying it onto a server so that people could use it.
JMR: What was it like working in different start-up environments and can you share the initial breakthrough moment in developing the technology behind Scholarcy?
The key thing about working for various start-ups is that until I did that, I never got the sense that I could build my start-up. I worked with Mendeley, BabylonHealth, and RefMe where I met my Co-Founder, Emma Warren-Jones in 2016. I learnt a lot of skills that helped me develop the idea for Scholarcy. It wasn’t simply a piece of software. There is a whole process that goes with it, like how are you going to realise this thing, how are people going to use it on the internet?
From designing something, building something, releasing it and then maintaining the software and updating it. It needs to be seamless. That is the technical infrastructure, which I learned while working as the NLP Lead with BabylonHealth.
By 2018, I had learnt enough that I figured I could finally build what I wished I’d had nearly ten years before during my PhD, so I left my full-time job, went freelance, and started building what became Scholarcy. I didn’t know exactly what a solution would look like, I just knew there was a problem. It wasn’t until I worked with RefMe that I became more familiar with app and Chrome extension engineering, although this was focused on reference management. It occurred to me that it would be useful for researchers to see more information than just the author and the abstract before deciding whether to read a paper in full. Why don’t we show the important informant around the paper or article, and what those key findings were or show the figures and tables? To be honest, we are still evolving that solution and it keeps changing. Over the last couple of years, we have stuck to the main principle: that our users have this separate, one-page view of the document that provides them a breakdown of the paper in sections, helping them to view it from different angles such as the top five highlights, comparative analysis, and related articles.
I suppose the key breakthrough moment for me was that if we could convert any document format into a universal structure, then I could build a machine learning model that could classify each line in the document according to its structural and semantic function, for example, this is the title, this is the abstract, this is a section heading, this is a reference, this is a key claim, this is an important point. If we could do that, then we could reinvent the document itself, make it more interactive, and slice and dice the information in many ways.
We could progressively enrich the document and reveal more information about the knowledge it contained.
That’s when I had the idea that we could turn a document into our interactive, summary flashcard format.
JMR: From a technical standpoint, what differentiates Scholarcy's AI from other academic tools available today?
Although we do use large language models, such as Mixtral, GPT-x, and others, we are not reliant on them, as we have built our own technology, which has quite modest hardware requirements.
Our tech is focused on automatically generating interactive summary cards from any document, so it is very specific, but it is what sets us apart.
Most other tools are focused on search and discovery, which is great, and necessary, but we are focused on the next step: transforming the reading and learning experience once you’ve discovered what you want to read. Many students have an assigned reading list, so search and discovery are perhaps less useful to them, what they do need to do is read, understand, apply, and synthesise the reading they have been given. It’s not just about the AI though, the whole user experience is important, and I want to recognise the fantastic work that our co-founder Emma has led on this, along with Jakub, John, Lisandro, Matias, Bruno, Karl, and Oliver.
JMR: How do you envision the role of AI in shaping future educational technologies, particularly in terms of personalising learning experiences?
It’s a difficult question to answer as people are much more concerned about privacy now than when, say Facebook, was new and everyone was sharing everything online. I’d first ask – do people want personalised learning experiences, and what kind of personalisation? How strong is the demand for an overseeing AI that will personalise content for you, and decide what and how you learn?
I think serendipity is so important – I’ve learnt so much from just browsing, online and in the real world, or following an off-topic train of thought in an article, picking up on something that others might find irrelevant, or seeing a random comment somewhere.
I’d like to see a way to embed serendipity somehow. It’s something we try to do at Scholarcy – we don’t provide recommendations, we don’t personalise, but we do automatically highlight things that we think look interesting.
Sometimes people say ‘This isn’t what I would have highlighted … but it’s led me to a new train of thought’ which is great.
Since the pandemic, people have had to learn how to be self-sufficient with their education. Lectures went online. Students were no longer attending lessons in person anymore. Perhaps, they weren’t getting as much support as they were hoping for, and students were turning to other services to help with their learning and gain independence.
JMR: Given Scholarcy's role in the domain of assistive technology and inclusive education, how has it been tailored or adapted to meet the specific requirements of neurodivergent students and researchers?
PG: Technically, we asked ourselves, what can we do to help all students? A company called, Assistive Solutions got in touch and informed us that Scholarcy was great for neurodivergent students because of the way it broke down long texts into bite-size chunks of information that didn’t overwhelm the user.
Scholarcy can help you thoroughly navigate complex information. I was building the software as I found it useful, and if I found it helpful then I thought maybe other students would too.
Last year, we hired Oliver Back, our Customer Support and Community Manager, and he has done a fantastic job organising our support channel and creating videos, training manuals, and webinars. His background was working as a Specialist Autism Mentor and Assistive Technology Trainer. Before joining our team, he trained people on how to use Scholarcy, who had ADHD, or autism. Oliver is good at telling us what users need or find useful when using the software.
We do not want to be seen as a tool that is beneficial for one demographic, we built a product that is universal and can help users who have dyslexia or learning difficulties. Scholarcy is made to break down information and present it in a more accessible way, and that can benefit everybody.
JMR: Can you elaborate on the technical challenges of scaling Scholarcy's services to accommodate a growing global user base?
PG: Our users process several hundred thousand, heading towards 1 million, documents per day on our platform.
So as our user base grows, one challenge is scaling our core engine to be able to handle this load and keep it responsive.
I met our Chief Technology Officer, Bruno Bonamin, when I was working at RefMe and we stayed in touch over the years. When I began working on Scholarcy, I needed his assistance in setting up the infrastructure. He has plenty of experience in building engineering teams and he has done a fantastic job in constantly improving how we work. We are going to have a much faster, responsive upload and processing experience soon because of his expertise.
JMR: What upcoming technologies or AI advancements are you most excited about incorporating into Scholarcy?
PG: We have some cool stuff coming around information visualisation, interactive mind maps, and showing the strength of evidence based on the quality of its reporting and analysis, plus how it compares to what has gone before.
The combination of the AI and the user interface, using the interactive flashcard format is what distinguishes Scholarcy that we did start by building our technology. We had a lot of different machine language models for specific tasks within a document. For example, we had a model that identifies when somebody is talking about how their work compares with previous research and that is what forms the basis of our comparative analysis section.
Our comparative analysis is unique and important. What does the author have to say? How does their body of work sit within the wider body of knowledge? Scholarcy classifies that information by identifying how the author builds on their argument or confirms their findings.
JMR: In leading a technology-driven startup, how do you foster a culture of innovation and continuous learning within your team?
PG: I think giving people access to whatever tools and learning they need, combined with the freedom to experiment and freedom to fail. One thing that used to frustrate me working for big companies was having to ask for permission to get a certain piece of software or to try a project and there were all these hoops you had to jump through.
Here, if someone has an idea for something new, I say, go and try it and report back.
Let me know what you need.
I like to think that we have a ‘no blame’ culture – if something goes wrong, we fix it and learn from it. I prefer to ‘ask for forgiveness, not for permission’.
JMR: Lastly, what advice would you give to entrepreneurs aspiring to make a significant impact in the EdTech sector?
PG: First, identify a problem that you have encountered, either in the past or now, that still hasn’t really been solved for you, and that you still care about. Chances are that others have this problem too.
You can then validate this by interviewing potential users first, to see what the demand might be, or you can do what I did and just start building something that addresses the problem, but the key thing I think is that you have good knowledge and experience of the pain point that you want to solve. Or if you don’t have it directly, care about it enough that you want to spend several years focusing on it.
Since building Scholarcy, other pain points have become apparent around the whole ecosystem of running an online business that has given me ideas for future products, as those pain points still haven’t been solved yet. So, there is always something new. Try to do something that others aren’t doing or haven’t thought of yet. Second, try not to go it alone, find a collaborator who shares your vision and enthusiasm but who brings something complementary, so if one of you has great tech skills, the other should be great at sales or marketing or building a team and coordinating. You’re going to be working closely for a long time so choose wisely!
Phil is the embodiment of a classic entrepreneur. His adaptability, and commitment to building and designing software that will benefit students from all over the world is truly commendable. Phil and his team are constantly working on bettering Scholarcy and creating a product that stands the test of time, by allowing students to become self-reliant.