Prot IV

Inspiration

As more startups rely on AI tools to write and modify code, we noticed a recurring point of friction. Automation moves faster than human oversight.

Simple mistakes like committing an API key are easy to fix. However, high-risk vulnerabilities require human intervention to better understand the problem and patch it up.

Thus, we built Prot IV to solve this. Prot IV answers the question of what an autonomous system should do when it doesn't have the confidence to act alone.

What it does

Our product continuously scans the codebase within the IDE to ensure there are no major security vulnerabilities. If any vulnerability is found, it is tested to determine the risk level. If this aforementioned risk level exceeds a threshold, it will notify an on-call security engineer and Prot IV will brief the engineer in a call. If the risk level is below this threshold, it will log the vulnerability and fix it on its own. An example of this is leaving an API key public; Prot IV will remove it and place it in the .env file. At this time, the codebases are simulated.

How we built it

ProtIV is a full-stack prototype with 3 sides:

A backend that scans and classifies vulnerabilities
A frontend that visualizes the state of the issue
A voice agent that creates and joins a live session to explain vulnerabilities and wait for human approval.

We utilized the Devswarm IDE to script several different iterations of the project, utilizing different Gen AI tools to prototype.

Challenges we ran into

There were several challenges the two of us faced. Integrating the LiveKit voice AI into our own product proved difficult, as we were struggling to keep the AI speaking English and even at many points make it speak at all. This was especially during prototyping the frontend. Coordinating this consistency was difficult.

Accomplishments that we're proud of

We build a solid prototype with a backend and frontend, while immersing ourselves in a new IDE! We completed our first hackathon with a functioning demo that aligns with what we wanted to show.

What we learned

Lots of iterating and prototyping. For my first hackathon, there's a lot more that goes into it than I realized. It also helped us think critically about when to use AI code generation tools ourselves for our project.