Clarity

NVIDIA TRACK

Inspiration

Engineers may often run into errors and struggle finding which errors to tackle first! We wanted to build a web app that can analyze a public github repository and provide solutions to assist the engineer in resolving the top open issues

What it does

Clarity retrieves a public github repository which then uses the Github REST API to fetch issues, PRs, comments, and reviews for the NVIDIA Nemotron API (Specifically the Nemotron Nano 9b-V2 Model) to analyze. It then produces a list of the top X issues with descriptions of the problem, ratings on severity, and how to fix it.

How we built it

Clarity is built using a multi-agent structure to divide tasks and communicate with each other to generate an analysis and a pseudocode solution to patch the issue

Challenges we ran into

After a project-pivot, we ran into the challenge of determining what model would work best for our new approach (Which we settled on Nano 9b-V2 for its superior reasoning). There were also issues with the formatting from the output that was returned by the code generator agent. In case the JSON parsing fails, we can extract raw code instead.

Accomplishments that we're proud of

We are proud of creating a web app despite the large setback with the project-pivot

What we learned

We learned how to utilize multiple APIs to communicate and work with each other. For most of us, this was our first time tackling this type of project. We also learned how different models work and what their best use-case is.

What's next for Clarity

In the future, Clarity should also provide the ability to identify the 'bottom' most issues for engineers who may not have much experience or want to start with smaller issues.