Bugzooka | Devpost

Inspiration

We've all spent time debugging websites and web apps; spending hours chasing miniscule bugs down tiny workflows instead of improving the main functionality of our projects. Manual QA is incredibly slow, and scripted tests can miss important real-world details. This enormously time-consuming and easily automatable process was one that caught our attention as developers working on our own apps -- leading to the birth of BugZooka.

What it does

BugZooka is your software team's determined, bug-hunting agent that stops at nothing to expose the errors in your websites and applications. Given a URL, BugZooka browses your app like a real human would; going through buttons, following links, filling forms, and triggering real edge cases. BugZooka can run on two different search options; "discovery mode," where it audits your given URL in an effort to expose hidden bugs, and "guided mode," where you can provide explicit instructions and (optionally) expected outputs for specific test cases. From there, it's all hands-off. Simply deploy your agent and expect a list of readily-made, in-app bug reports -- complete with screenshots, bug descriptions, and recreation steps. Once you're ready, confirm the reports and automatically create Jira tickets or GitHub issues to bazooka the bugs!

How we built it

BugZooka was built in just 24 hours using cutting-edge AI tools, sponsor APIs, and an unhealthy amount of caffeine. At its core, BugZooka runs on an autonomous testing agent orchestrated by Agentuity and powered by Gemini's brilliant API.

Gemini acts as the eyes and brain of BugZooka, visually analyzing each web page, recognizing buttons, links, and input fields; interpreting the page structure the same way a human tester would. As our agent explores, Gemini intelligently identifies potential bugs, takes contextual screenshots, and provides detailed observations about what it sees and how the UI responds.

Agentuity provides the backbone that keeps our agent alive and organized. It handles task scheduling, execution tracking, and real-time communication between the frontend and backend; acting as the vessel through which Gemini executes.

Our full tech stack included: Frontend: Next.js + Tailwind CSS for a fast, clean, and interactive dashboard. Backend: FastAPI for orchestrating agent runs, handling results, and serving reports. Agent Engine: Selenium integrated with the Gemini API for intelligent visual navigation and interaction. Orchestration: Agentuity for deployment, lifecycle management, and event coordination. Authentication: Auth0 Integrations: One-click issue creation through Jira and GitHub APIs.

Challenges we ran into

While we faced endless challenges with basically every part of our project (as Hackathon rookies), the biggest technical challenge we ran into was deploying our agent. Creating an autonomous tester that acts like a real human QA is no easy feat, and took tons of testing and iteration before we could pull it off.

However, this was the first hackathon for everyone on our team, so just creating a full-stack application from end-to-end (let alone in 24 hours) was perhaps the biggest challenge. It took a huge amount of effort, but was incredibly rewarding and a very valuable learning experience.

Accomplishments that we're proud of

While we are incredibly proud of the fact that we managed to create a genuinely useful working agent in 24 hours, we are even more proud of our work ethic and ability to solve the numerous difficult problems we faced as Hackathon first-timers. We faced issues everywhere, everything from Auth0 to dumb Git decisions to spilling Celsius all over our keyboards. Nonetheless, we pushed through and created an awesome product. We're proud of that!

What we learned

First and foremost, we learned how to take an ambitious idea and turn it to reality in 24 hours. Coordinating merge conflicts, falling-apart frontend components, and rogue agents while fighting a strong case of the munchies at 3AM forced us to think fast, prioritize, and collaborate efficiently.

From a technical aspect, we explored the limits of modern AIs. We learned that Gemini, while incredibly brilliant, required carefully designed prompts, context management, and fallback logic to perform as we intended

From a product perspective, we learned that user experience matters just as much as technical brilliance. A clean dashboard, clear results, and fun details like roast messages made the product feel alive and approachable.

Most importantly, we learned the importance of collaboration and team energy. Splitting up work accordingly, learning to compromise, playing on each other's strengths, and remaining optimistic when all seemed lost was perhaps the most important thing that kept BugZooka alive.

What's next for Bugzooka

As developers and software people ourselves, we feel that BugZooka has immense potential. A 24-hour hackathon -- while a great place to get started -- simply isn't enough to truly solve all the problems we're capable of. Next up, a full-fledged autonomous QA platform that expands beyond basic bug discovery towards mobile and desktop app support, API testing and validation, and testing of hundreds of pages at once. Long term, the goal is pretty simple; bazooka the bugs!