Inspiration 😭

It was 2:40 AM on a school night. My team and I were staring blankly at a screen full of thousands of lines of unhandled runtime exceptions and cryptic stack traces. We had spent the last four hours trying to figure out why our project kept crashing, only to realize... we had spent three and a half of those hours just trying to find which line of code actually triggered the cascade.

Between juggling heavy course loads, back-to-back lectures, late-night study sessions, and severe sleep deprivation, human error is inevitable. But when production breaks or a project deadline is minutes away, panic sets in. Browsing through dozens of Stack Overflow threads while squinting at raw terminal logs is exhausting, inefficient, and demoralizing.

We realized that developers don't fail because they can't code—they fail because they lose critical time and energy context-switching and panicking over indecipherable error dumps. We built PanicTriage so no developer ever has to spend their late nights deciphering a wall of red text alone.

What It Does 🚀

PanicTriage is an intelligent, emergency-response AI agent for developers in crisis.

When your application crashes, you simply paste your raw stack trace or terminal output into PanicTriage. In seconds, the agent:

Isolates the Root Cause: Instantly extracts the exact file, line number, and core fault—cutting through the noise of nested call stacks.
Generates an Actionable Fix: Delivers a ready-to-use code snippet tailored to your exact runtime error.
Explains It Simply: Provides a high-level breakdown of why the bug occurred so you learn and avoid repeating it.

How We Built It 🛠️

Frontend: Clean, high-contrast, distraction-free UI designed for high-stress environments.
Backend: Fast API gateway that ingests unstructured error streams.
AI Engine: Prompt-engineered LLM pipelines trained on structured stack trace parsing and diagnostic JSON outputs to deliver precise, hallucination-free fixes.

Challenges We Ran Into 🤯

Parsing unformatted log outputs from different languages (Python, JavaScript/TypeScript, C++) was tough because stack traces vary wildly in layout. We had to carefully structure our system prompts to enforce strict JSON schemas across diverse crash dumps without losing critical debugging context.

Accomplishments That We're Proud Of 🎉

Building a functional, end-to-end agent in under 6 hours!
Successfully reducing raw, 200-line stack traces into a single-line diagnosis and fix within seconds.

What We Learned 💡

How to design LLM context windows specifically optimized for code diagnostics.
That speed and UX matter most when a user is in panic mode—clarity trumps complex UI every time.

What's Next for PanicTriage 🔮

CLI Integration: Running panic-triage run main.py directly in the terminal to automatically catch and diagnose unhandled exceptions in real time.
IDE Extension: A VS Code extension that pops up a floating diagnostic panel whenever a local build fails.