Edge AI ASL Recognition on Raspberry Pi 5 – Looking for Feedback on My System Design [P]

I'm implementing an offline ASL recognition system on Raspberry Pi 5 using MediaPipe hand landmarks and TensorFlow Lite. The system recognizes the ASL alphabet and converts it to text and speech without an internet connection. My current pipeline is: MediaPipe (21 hand landmarks) Landmark normalization TensorFlow Lite model on Raspberry Pi 5 OLED display + offline TTS I'm trying to decide between a 1D CNN, MLP, or GRU for landmark-based classification. My priority is low latency and efficient edge deploymen