Perdus — Project Story
Inspiration
Traditional lost-and-found systems rely on public item listings and manual searches. This makes recovery slow and encourages false claims. We wanted a system that works like verification instead of browsing: users describe what they lost, and the system privately checks if a match exists.
What it does
Perdus is a privacy-first lost-item recovery platform. Users submit inquiries using text or photos. The system compares them against a hidden inventory and returns high-confidence matches only after assistant review and ownership verification.
How we built it
We built two connected web apps:
- A Public App for submitting and tracking inquiries
- An Assistant App for managing inventory and reviewing matches
Tech stack:
- Frontend: Next.js + ShadCN UI
- Backend: API for inquiries, matching, and authentication
- Data: Private catalog accessible only to authorized assistants
Matches are ranked and filtered before review.
Challenges we ran into
- Designing a flow where the inventory is never publicly visible
- Reducing false positives when many similar items exist
- Creating ownership verification without exposing sensitive details
- Managing role-based access for users vs assistants
Accomplishments that we're proud of
- Built an end-to-end matching system
- Implemented confidence scoring instead of keyword search
- Added adaptive follow-up questions when too many matches exist
- Designed ownership verification based on item details
- Maintained strict separation between public and private data
What we learned
- Privacy must be part of the core design
- Simple attributes become powerful when combined
- Human review improves trust and accuracy
- Fraud prevention is essential for recovery systems
What's next for Perdus
- AI-based image similarity
- Duplicate and fraud detection
- Integration with institutional systems (schools, transit, airports)
- Stronger encryption for item data
Perdus — Find what’s lost without exposing what’s found.
Log in or sign up for Devpost to join the conversation.