1st Place · HackMX 6th Edition
IdentifEye
Snap a photo of any product and get the closest match from Liverpool's catalog, plus LLM-powered recommendations.
- My role
- Computer vision & AI
- Team
- 5 people
- Event
- HackMX 6th Edition · Tecnológico de Monterrey
- Year
- 2025
The problem
Liverpool asked teams to let customers find a product with nothing but a photo. Text search fails when you don't know what something is called, and the catalog was large and full of near-duplicate images.
What I built
- Image-embedding search on Vertex AI that matches any photo to catalog items.
- Recommendations for related products using GPT-4 Vision.
- Embedding-based duplicate removal for a catalog of any size.
- Concurrent microservices on Google Cloud, so one failure never takes down the web app.
- Live embedding storage in a database to optimize resources.
- React + Vite front end hosted on Firebase and Google Cloud Run.
Results
- 1st place in the El Puerto de Liverpool AI Challenge at HackMX VI, out of 100+ competing teams.
- Liverpool publicly congratulated the team on LinkedIn.