Before founding Unitary, I completed a PhD in mathematics at the University of Cambridge, where I tried to answer the question: what happens to objects when they fall into a black hole? I was always interested in understanding and solving complex problems using mathematics – but I also wanted to do work that made a real difference. That’s why I moved out of academia and into the world of startups. To make that change, I joined Entrepreneur First, where I met my co-founder and former Meta AI researcher James [Thewlis] and together we founded Unitary, driven by an ambition to make the internet a safer, more transparent place.
Tell me about the business – what it is, what it aims to achieve, who you work with, how you reach customers and so on?
Unitary is building contextual content moderation solutions by using machine learning models to analyse visuals, text and audio together to interpret content in context. Our systems understand the meaning behind online images and videos to detect harmful content and ensure that content moderation is fairer and more accurate. Context is often critical for understanding: nudity, for example, in the context of a medical diagnosis carries very different significance to nudity in a sexual context. With our contextual AI models, Unitary aims to understand the visual internet and build a safe and transparent digital space. We are already incorporating meaning into the analysis of 2.3 billion frames from social media videos every day.
How has the business evolved since its launch? When was this?
Since our founding in 2019, we’ve grown our team to over 20 people in six countries, taking a mere idea and turning it into a live product with numerous partners. We have developed solutions across text, image and video, refining our value proposition over time and scaling our products so that we can process millions of images and videos every day. In addition to commercial partners, we have also worked with non-profits and have been involved in the proactive detection of hate speech surrounding election campaigns in Africa. We also released open-source tools such as our text-based toxicity detection model Detoxify. In July 2022 alone, Detoxify was downloaded almost two million times. It has been exhilarating to see Unitary grow and to be able to make a tangible difference to online safety.
How are you funded?
In 2020, we closed a £1.3M seed funding round led by Rocket Internet’s GFC, who were joined by January Ventures, SGH Capital and a number of angel investors. We’re excited to share an update on our progress in the coming months.
What has been your biggest challenge so far and how have you overcome this?
As with most companies, the onset of the pandemic forced us to adapt to remote working and re-think many of our practices and processes. We then had to scale the team in a completely remote environment and navigate the challenges of creating a strong company culture and well-gelled team virtually. Despite the initial challenges, we really embraced remote working and by being more proactive and deliberate about building meaningful connections among the team, we realised that there are so many benefits to this new set-up. We took the decision to become a remote-first company, which enabled us to hire great team members all over the world. Remote working is now a core part of who we are, and we now can’t imagine existing any other way.
How does Unitary answer an unmet need?
Between 2020 and 2025, the amount of information on the internet is set to increase by a factor of ten. This trend is driven by video, which makes up over 80% of traffic and is concentrated on social networks and platforms. Given the huge scale and complexity of the moderation challenge, human reviewers cannot and should not have to tackle this alone – particularly in light of the emotional toll that human reviewers often experience owing to the repeated exposure to traumatic content. We believe that effective content moderation needs to include both human reviewers and automated solutions to keep up with the pace and scale of the challenge. That is why we are developing machine learning models that take into account multiple different sources of information before classifying a particular piece of content. For example, when classifying a social media post, we take into account the title, comments, and visual data to offer a richer understanding of content by determining the meaning behind media.
What’s in store for the future?
Our long term goal is to understand the visual internet and build a transparent digital space. The first step to this is understanding what content exists on the internet: we are working to perfect the process of content categorisation, which in turn will offer the potential to empower people to directly determine the type of content they see online.
What one piece of advice would you give other founders or future founders?
Don’t give up! Building a company is hard and, in my experience so far, it never seems to get easier. But the type of challenge is constantly evolving, so getting used to fire-fighting and quickly adapting to ever-changing situations is the name of the game.
Sasha Haco is CEO and cofounder of Unitary.