In recent years, GitHub has fundamentally changed developer workflows. By centralizing code on an easily accessible platform, the company was able to rapidly change the way people code. Following in these footsteps, Israeli startup Codota wants to further optimize workflows for the often neglected developer community — this time with machine intelligence. The company is announcing a $2 million seed round from Khosla Ventures for its autocomplete tool that helps engineers push better code in less time.
Codota interfaces with integrated development environments like Eclipse, expanding on intelligent code completion. Instead of just offering up brief suggestions of intended code, Codota can recommend larger chunks.
Co-founders Dror Weiss and Eran Yahav took advantage of open source code on the internet from GitHub and StackOverflow to build Codota. All of this public code was fed into machine learning models to enable them to recognize higher-level meaning across blocks of code.
Programing languages share a lot of structural similarities with their distant spoken cousins. Words can be arranged in infinitely many ways to express a single thought or sentiment. Likewise, the same command can be represented in code in a number of ways. This is why it’s so critical that Codota understands the macro picture of what code is doing.
Of course natural language and code are not completely analogous. The team explained in an interview that in natural language processing, meaning is determined by looking at nearby words. Programs are more structured and meaning isn’t always strongly correlated with locality. So instead of just training on text, Codota also focused on the behaviors of a program.
Aside from improving speed and accuracy, Codota can help with discovery and education. Because Codota has been trained on millions of API implementations, it can help offer up best practices to developers. When open side-by-side with an IDE, the tool can highlight irregularities and demonstrate better ways to write code, lessons often pulled straight from the original creators of libraries.
The startup makes its money by allowing enterprises to keep their internal code private while benefitting from Codota’s insights. Right now the tool is limited to Java, but in the future additional languages will be added.