DeepCode will find errors in the source code of the software using AI

Swiss startup today deep code, which uses artificial intelligence and machine learning to automate code analysis, announced a $4 million investment from venture capital funds Earlybird, 3VC and Btov Partners. The company plans to use these funds to implement support for new programming languages ​​in its service, as well as marketing the product in the global IT market.

Code analysis is essential to detect bugs, potential vulnerabilities, formatting violations, and more in the early stages of software development before the code is used anywhere. Usually this procedure is carried out in parallel with the development of new code and immediately after it is completed, preceding the stage of testing itself. “Software testing looks at the code from the outside, but code analysis allows you to look at it from the inside,” explains DeepCode co-founder and CEO Boris Paskalev in an interview with VentureBeat.

Most often, code review is performed by code authors in conjunction with colleagues and managers to identify obvious errors before moving on to the next stages of development. And the larger the project, the more lines of code need to be checked, which takes up a significant amount of programmers' time. Tools that should speed up this process have been around for a long time, such as static code analyzers such as Coverity and PVS-Studio, but they are usually limited in their capabilities, as they focus on "annoying and repetitive stylistic problems, formatting and small logical errors,” explains Paskalev.

DeepCode, in turn, covers a wider range of problems, for example, by detecting vulnerabilities such as opportunities for cross-site scripting and SQL injection, since the algorithms embedded in it do not just analyze the code as a set of characters, but try to understand the meaning and purpose of the work written programs. At the heart of this is a machine learning system that uses billions of lines of code from public open source projects for its training. DeepCode analyzes previous versions of the code and subsequent changes that have been made to it in order to study what errors and how real programmers fixed in their work, and then offer similar solutions to their users. In addition, the system also uses traditional prediction algorithms to find possible problems in the code, like the static analyzers mentioned above.

One of the key questions when using DeepCode is how reliable is automatic code checking? An analysis accuracy of less than 100% means that developers will still have to analyze their code manually. In that case, how much time would it actually free up using tools to automate this task? According to Paskalev, DeepCode will be able to save developers about 50% of the time they currently spend searching for bugs on their own, which is quite a significant figure.

Developers can connect DeepCode to their GitHub or Bitbucket accounts, and the tool also supports local GitLab configurations. Additionally, the project has a special API that allows developers to integrate DeepCode into their own development systems. Once connected to the repository, DeepCode will analyze each code change and flag potential issues.

“On average, developers spend about 30% of their time finding and fixing bugs, but DeepCode can save half that time now, and even more in the future,” says Boris. “Because DeepCode learns directly from the global developer community, it is able to uncover more issues than a single person or group of reviewers could ever find.”

In addition to today's investment news, DeepCode also announced a new value policy for its product. Until now, DeepCode has only been free for open source software development projects. Now it will be free to use for any educational purpose and even for commercial companies with less than 30 developers. Obviously, with this step, the creators of DeepCode want to make their product more popular with small teams. Additionally, DeepCode charges $20 per developer per month for cloud deployments and $50 per developer for local support.

The DeepCode team has previously received a $1 million investment. With another 4 million, the company said it plans to expand the programming languages ​​it supports beyond Java, JavaScript and Python, including adding support for C#, PHP and C/C++. They also confirmed that they are working on their own IDE.

Source: 3dnews.ru