Inspiration
International shipping is surprisingly fragile. A package can travel thousands of miles without issue, only to be delayed because of an incorrect customs declaration, a missing compliance label, or a mismatch between what's actually inside the box and what's written on the outside. We wanted to build something that catches those mistakes before the package ever leaves the seller's desk.
Our goal was to create an affordable AI-powered inspection system that helps small merchants verify their shipments, reducing costly customs delays and making cross-border shipping more reliable.
What it Does
Our prototype uses two cameras working together.
- A top-mounted camera looks inside the open package and identifies its contents using computer vision.
- A side-mounted camera scans the exterior of the box as it rotates on a turntable, reading shipping labels, barcodes, and compliance stickers with OCR.
The AI combines information from both viewpoints to verify that the package contents match the attached labels. If something is missing, incorrect, or inconsistent, the system alerts the user before the package is sealed and shipped.
How We Built It
We built the system using computer vision, OCR, and AI reasoning running on inexpensive hardware. The cameras continuously capture the contents and exterior of the parcel while a motorized turntable rotates the box, allowing the side camera to inspect every face without requiring multiple cameras.
The software fuses information from both camera feeds, extracts text from labels, detects objects inside the parcel, and checks whether everything is consistent before giving a pass or warning to the user.
Challenges We Ran Into
Like many hackathon projects, the final prototype is... let's call it battle-tested.
Our build is admittedly pretty scrappy. During development, one of our drive motors failed rather spectacularly and essentially destroyed itself, forcing us to redesign parts of the mechanical system with the components we still had available. We also had to improvise mounts, wiring, and structural pieces after several parts didn't arrive on time or failed under testing.
Getting two cameras synchronized with a rotating platform also turned out to be far more difficult than expected. Small timing differences caused blurry scans or inconsistent OCR results, so we spent a significant amount of time tuning the rotation speed, camera timing, and software pipeline to make the inspection process reliable.
Despite the hardware setbacks, we focused on demonstrating the core idea: affordable AI-assisted parcel inspection that combines multiple viewpoints to catch shipping mistakes before they become expensive customs problems.
What We Learned
This project taught us that building intelligent hardware is just as much about engineering around failures as it is about writing code. Computer vision models are only one piece of the puzzle. Mechanical reliability, lighting, synchronization, and hardware integration all have a huge impact on whether an AI system actually works in the real world.
Most importantly, we learned that even with limited time, broken hardware, and a very improvised prototype, it's possible to build something that demonstrates a real solution to a real problem.
Log in or sign up for Devpost to join the conversation.