On-device ML for smart glasses (ESP32-S3): OV2640 → TFLM/ESP-NN (int8) → BLE.
Low-power modes, OTA/DFU, and KPI tooling for latency/FPS/mW.

• On-device inference with TensorFlow Lite Micro (TFLM) / ESP-NN (int8)
• Camera pipeline: OV2640 → I2S/DMA → PSRAM double buffer → preproc → embeddings → BLE
• Low-power: adaptive framerate, duty cycling, sleep/retention; BLE interval tuning
• OTA/DFU with crash-safe rollback
• KPIs: latency / FPS / mW dashboards; Python benches (pyserial/pyvisa + SMU)

1. Install ESP-IDF v5.x and Python 3.10+
2. idf.py set-target esp32s3
3. Configure pins and BLE params in firmware/main/app_config.h
4. Build/flash/monitor: idf.py build flash monitor

Clean-room demo inspired by OpenGlass; don’t copy code verbatim from upstream.