2025 · Hardware / Robotics
Basil Home AI — Smart Refrigerator Inventory Engine
QNX RTOS + Raspberry Pi camera, custom YOLOv5 grocery model, FastAPI inventory, Gemini recipes.
Hack the 6ix Winner · Deloitte Best Use of AI for Green
The problem
Household food waste is a visibility problem: groceries get pushed to the back of the fridge, forgotten, and expire before anyone cooks them — and a smart-fridge retrofit usually means a $2,500+ appliance, not a cheap add-on. The people who waste the most (overworked families, seniors) can't justify that upgrade.
The constraint
Live capture inside a fridge needs an OS that wakes instantly and schedules captured camera work deterministically — desktop Linux boots too slowly and jitters. And the real challenge for a team of four with different stacks: camera node, vision model, backend, and a mobile app the group barely knew all had to integrate in roughly 36 hours.
What I built
I configured BlackBerry QNX RTOS on a Raspberry Pi as the fridge brain, standing up networking + a lightweight server so the camera streamed out of the appliance in real time. The team trained a custom YOLOv5 model on 300+ hand-curated food images; the CV script (torch.hub, ~0.8 conf) counted grocery classes per frame. A FastAPI backend owned inventory state and, importantly for the demo, *tagged how soon each item expires*, splitting the fridge into USE-TODAY / THIS-WEEK / LATER buckets. Gemini then drafted recipes from the *soonest-expiring* items, and an Expo app made the whole thing hands-free with AssemblyAI voice. National data (58% of Canadian food wasted, $1,300+/yr per household) set the stakes in the pitch.
The outcome
Won Deloitte's "Best Use of AI for Green" at Hack the 6ix 2025 with a live fridge-camera-to-inventory-to-recipes demo. The same brains later became Basil's production Android app + FastAPI backend.