Edge devices often can't afford to ship raw sensor data over the network.
This project trains an autoencoder to compress and reconstruct sensor
readings, tuning the latent-vector size for the smallest representation that
still reconstructs cleanly.The encoder was converted to run on-device with TensorFlow Lite Micro
(microcontroller-side inference), while the decoder lives behind a Flask API
on AWS EC2, with S3 handling encoded payload storage.
Highlights
Autoencoder tuned for compact latent representation of sensor data
Microcontroller-side inference via TensorFlow Lite Micro
Decoder exposed as a Flask API on AWS EC2 with S3 integration
