The AI Baby Monitor operates by capturing video stream frames with a script called `stream_to_redis.py`, which sends the frames to Redis streams. Another script, `run_watcher.py`, retrieves the latest frames, combines them with the safety rules, and processes them through a local vLLM server running Qwen 2.5 VL. The model returns a structured JSON with fields like `should_alert`, `reasoning`, and `awareness_level`. If an alert is necessary, a beep sound is played. Additionally, a Streamlit page displays both the camera feed and the logs from the language model, providing a comprehensive monitoring interface.
Key takeaways:
- AI Baby Monitor uses a stack of Redis, vLLM, and Streamlit to monitor video streams and safety rules.
- The system alerts parents with a beep sound if a safety rule is broken, allowing them to check on their baby.
- The tool is designed to be an additional set of eyes, not a replacement for adult supervision.
- Streamlit displays both the camera feed and logs from the language model for transparency.