Herve Vu Roussel and his team want to solve one huge problem that most software engineering teams face: code collaboration.
The issue is this: imagine having 40 editors working on the same document at the same time. It’s almost a certainty that your shared word processor platform would crash, with the document itself turned into a mess.
Now, imagine that the document has 5 million lines of text. It’s 70 times the length of Leo Tolstoy’s War and Peace. That’s what software engineers have to deal with when they collaborate.
Roussel founded QuodAI in 2018 to apply deep learning and natural language processing to convert raw source code into plain English. The platform has also already developed an AI assistant for software engineer teams, which helps teams with their onboarding process for newly hired software engineers and collaborating on a daily basis.
Increased complexity of software engineering
Software engineering is at the heart of today’s innovation. The race to build the next product and app makes it critical for companies to search for and recruit talent, as software engineers are constantly in high demand. Not surprisingly, software is also the industry with the highest turnover rate, according to LinkedIn’s data on half a billion professionals in 2018.
“Software engineering has become incredibly complicated over the last 15 years,” said Roussel. “Fifteen years ago, we had about six months to one year to plan the release. Today, we release every day, every week. The expectations of customers are also incredibly higher. You have to release for PC, Mac, mobile, watch, tablet—all different devices.”
During stints as CTO and head of data engineering for various companies, Roussel said he noticed the problem faced by startups when scaling their software engineering team at the growth phase. Companies have to add more software engineers to the team, which slows down the development process exponentially with every new hire, line of code, and feature.
While most companies probably do not match the scale of Google’s repository of 2 billion lines of code, it’s definitely a challenge to keep up with a growing codebase once you’ve started to add new features and have new releases.
In 2018, the idea for QuodAI started to materialize when Roussel met with data scientist Mikhail Filippov from Nanyang Technological University at the Entrepreneur First accelerator program in Singapore. Both wanted to leverage AI to help software engineers collaborate more efficiently. Dr. Filippov is now QuodAI’s co-founder and chief scientist.
Half a billion lines of code
Registered in Singapore and with engineering teams in both Singapore and Ho Chi Minh City, QuodAI wants to develop and streamline its own in-house AI technologies. The startup claims that it has developed AI models that have been trained on over half a billion lines of code from both open and proprietary source code sets.
QuodAI targets companies that “have problems scaling their engineer team.” Newly hired engineers can find their fit quickly because the codebase has been turned into natural language, which is searchable for every software engineer. This also solves the problem of coders writing in different programming languages.
For example, the AI assistant provided by QuodAI can tell a software engineer where to look for a chunk of code that logs in a user or forms a user interface panel. The system is also integrated with GitHub, the world’s largest social coding platform.
“We believe that training an AI that can understand source code is the key to expanding into other interesting areas like automated code completion, code review, and bug-fixing,” Roussel explained. “Building an AI company is not easy and how much you use AI is not black or white. The mindset for an AI company is very different. We talk about our product in terms of probability or confidence scores over deterministic actions.”
QuodAI charges a subscription fee per user per month, and offers both on-premise (as in disconnected from the internet) and hosted solutions for companies. Roussel said QuodAI has attracted enterprise customers including one unicorn and fast-growing startups. Like other software development and collaboration platforms such as Github and Atlassian, clients have agreed to give the startup access to their source code in order to translate and train the AI assistant.
As with other AI startups, the biggest challenge, Roussel said, is in acquiring enough data to train the AI system to translate code more accurately.
The startup previously participated in accelerator programs run by 500 Startups and Singapore’s Entrepreneur First. It is open to discussion about raising additional funding “given the exciting early output of our AI models.” The company has plans to expand to the US next year. “Software engineers are one of the most valuable resources in the world, and everyone is fighting for them,” Roussel said. “We’re solving a worldwide problem.”
This article is part of KrASIA’s “Startup Stories” series, where the writers of KrASIA speak with founders of tech companies in South and Southeast Asia.