HawkEye: 12 Hours. Zero Code to Execution.
Yesterday at the CSHub Local Hack Day, my first hackathon ever, my teammate Soroush Baraouf and I challenged ourselves to build a fully functional product in just one day.
The Result: Meet "HawkEye" 🚀
Resellers spend hours manually logging inventory. HawkEye changes that. You simply record a video pan of a clothing rack, and our AI analyzes the footage. It identifies items (e.g., "Vintage Levi's 501"), detects specific damage like a missing button, and instantly estimates market value.
My project role:
- Frontend & UX: Designed the user interface and overall experience.
The Challenge: Raw Video Processing
The main technical hurdle was processing raw video for the #GoogleGemini 2.5 Flash API.
The Fix: I built an #FFmpeg pipeline to extract frames and timestamps, enabling Gemini to process video at a granular level for deeper analysis. This unlocked a full spectrum of Gemini’s multimodal capabilities: video understanding, image interpretation, voice detection, text generation, reasoning, and real-time data retrieval.
How Gemini Powers the Backend
- Analyze video frames and identify items with high accuracy
- Detect condition details like missing buttons or wear
- Interpret audio cues (if present)
- Search the web for live market prices
- Reason about value ranges and suggest a lower and higher price based on condition
- Generate optimized listing descriptions for online marketplaces
We designed the pipeline so Gemini acts as the backbone of our product and an end-to-end assistant, from visual understanding to pricing intelligence to marketplace-ready text.
Deployment & The "Meta" Twist
The application was containerized with Docker and deployed on Render.
But here is the "Meta" twist: We didn't just use AI; we built with it. We utilized the #GeminiCLI (my favorite new AI toy) as an agentic assistant to handle boilerplate code and rapid debugging, allowing us to hit the strict 12-hour deadline.
A huge thanks to my teammate Soroush Baraouf for the incredible collaboration. Going from zero code to execution in 12 hours was an intense but rewarding experience.