Show HN: AI Catflap Prey Detector (github.com)

Catflap Prey Detector is an open-source, DIY edge-AI system that watches your catflap, detects when a cat is carrying prey, and temporarily blocks the flap by toggling an RFID jammer. The pipeline uses a Picamera 3 IR feed (24/7 night vision), YOLO11n (NCNN) for object detection/tracking, and a custom “Prey Detection API” to classify whether a cat is carrying prey; when prey is detected a 134.2 kHz FDX‑B RFID jammer is triggered via a 5V relay to keep the gift outside. Notifications, manual controls (lock/unlock/status/photo/ping) and alerts are handled through a Telegram bot. The project targets Raspberry Pi (Pi 5 recommended but ~1GB RAM sufficient), runs on Python 3.11+, installs with make/uv, and supports optional Google Cloud Storage for long-term image persistence. Required env vars include BOT_TOKEN, GROUP_ID and PREY_DETECTOR_API_KEY; the repo also documents wiring, systemd deployment, and testing. For the AI/ML community this is a compact example of real-world edge inference and actuator integration: lightweight YOLO inference on NCNN, a modular API-based classifier, and an easy path to swap in a self-hosted model by replacing detect_prey(). It highlights tradeoffs between cloud-hosted model APIs and local inference, data-collection opportunities for future model training, and practical concerns around latency, power and thermal management on Pi hardware. Note: the design physically interferes with an RFID device—check safety and local regulations before deploying.