As a team of developers, we are all too familiar with how tedious it can be to find and fix problems in code. The available code review/static code analysis tools did not satisfy our and many others developers’ needs. These tools create noise by flagging false positives and are only able to detect simple problems, mostly related to style and syntax. Therefore, we decided to build Metabob, a tool that speeds up debugging and refatoring code by up to 60%.
Metabob is an AI code review tool that automatically detects complex problems in code bases. Examples of these include problems such as unhandled edge cases, memory leaks, GPU/CPU runtime initialization errors, and hundreds more. Doing so requires a contextual understanding of the code base, which is enabled in step 1 through a graph-attention mechanism. In step 2, after a problem is found, Metabob passes the problematic code region along with enriched code context to a generative AI model to receive context-sensitive problem descriptions and code recommendations (the user is able to opt out of step 2). Metabob also allows users to interact with a generative AI by asking questions regarding the problem descriptions and code recommendations. Additionally, users can provide the generative AI with more context if they wish to improve the accuracy of the generated solution.
Features:
✅ Problem detection
✅ Generation of problem descriptions
✅ Generation of code recommendations to fix detected problems
✅ Ability to interact with a generative AI
Currently Metabob's AI is trained on analyzing Python code. Soon, we will expand to Java and Javascript and will add more programming languages at a later date.
