Inspiration

Iron Man's Jarvis is great for movies and holograms, but what if you wanted to make real-life productivity better. As a team of people who have been coding for years both in industry and academia, we know that the #1 use of a coder's time is fixing bugs. Almost 90% of the bug-fixing process is repeatable however. That's why we decided to teach an AI assistant to observe you as you code and help find the best solution to your bugs.

What it does

Jarvis is available at your service whether it's making navigation easier or helping you with coding issues! Come check us out at table 7! All you have to do is import jarvis at the top of your code and it will automatically start watching your code for any errors/bugs. It can either automatically compile/run your code at certain intervals or it can wait until you run the code yourself. If you see that your code has errored, you can say "Jarvis, look up this error" at which point it will take you to the most relevant sight for your issue.

Challenges

The two biggest challenges were making this work seamlessly in a common developer environment and producing relevant search results. The challenge with the seamless workflow is piping the errors to both the user and Jarvis at the same time with as little extra code for the developer as possible. The most ideal case is the developer just having to include an import statement. The reason for this is most developers won't use a tool if it's too much work to set up.

The other challenge is to find relevant results. Often times, Google search will pull up the most relevant answers. Other times, you have to comb through to the 5th or 6th page. We had to write a ranking algorithm that can order the links properly so that the coder gets the most relevant response.

Accomplishments we're proud of

Figuring out the seamless workflow was huge for us. We were able to hack together a way in which the error messages are both printed out for the user and copied in the right format for Jarvis. While it took several redirects and complicated piping methods, using Jarvis is extremely simple because of this hack.

Providing relevant results to coders. The best test was to use it ourselves as we coded the final portion of Jarvis to see if they were relevant. Of the 46 errors that we ran into, 38 were resolved through Jarvis's recommendation. That's 83%. If a developer saves on average 2 minutes on a bug that gets instantly resolved through Jarvis, that means I saved almost an hour and a half, which is huge for coders.

What we learned

Lower level processes relating to STDERR Pytorch basics to create a bug classifier

Share this project:
×

Updates