SafeLock

Photo detection circuit

Inspiration

We wanted a simple, fail-safe way to deter unauthorized access to a box/door (e.g., dorm storage, lab cabinet, shared equipment). If the system can’t confidently verify the authorized user, it should default to “deny” and alert.

What it does

SafeLock is a real-time safety guard:

A Raspberry Pi sensor detects a potential open/access event (light-based trigger in our demo).
The Pi requests a quick face verification from an auth server (laptop webcam + Gemini vision).
The system returns a strict true/false decision:
- AUTH_SUCCESS → buzzer stays off
- AUTH_FAIL (unknown / not logged in / unsure) → buzzer alarm
It’s designed to fail safe: when in doubt, it alerts.

How we built it

Python for both server + device logic.
OpenCV to capture and encode webcam images.
Gemini API (google-genai) to compare the authorized reference photo vs the live photo and output JSON.
TCP sockets for Pi ↔ laptop communication.
GPIO on the Raspberry Pi to read the sensor and drive the buzzer.

Challenges

Rate limits and reliability: added retry/backoff and stricter decisions (only approve when confidence is high).
Real-time behavior: making sure the loop is responsive without spamming API calls.
Networking and setup: ensuring the Pi and server communicate reliably on the same network.

What we learned

Building a clean “sense → verify → act” pipeline with a fail-safe default.
Integrating hardware triggers with networked AI verification.
Debugging real-time loops and socket communication under hackathon time pressure.