TINATSEIPORTFOLIO '26
Back to Projects
ML / Edge

Ripple

Wi-Fi Based Presence Detection · January 2026 – April 2026

Raspberry Pi 4BWi-Fi CSISTFTTensorFlowCNN-GRUFastAPIReact
Ripple

01 // THE PROBLEM

Camera and wearable occupancy sensing leaks identity. This project needed human presence detection that stays on-device and never records a face.

02 // THE SOLUTION

Built a privacy-centric presence system on Raspberry Pi 4B using Wi-Fi Channel State Information. Raw CSI becomes STFT spectrograms with center cropping to drop subcarrier noise, then a Hybrid CNN-GRU in TensorFlow classifies micro-movements at 96.4% validation accuracy. FastAPI streams occupancy JSON to a React dashboard.

03 // DEEP DIVE

Personal project

January 2026 – April 2026

What shipped

  • --Privacy-centric human presence detection using Wi-Fi Channel State Information captured on Raspberry Pi 4B hardware.
  • --Signal processing pipeline that turns raw CSI packets into spectrograms via Short-Time Fourier Transforms, with center cropping to filter subcarrier interference.
  • --Hybrid CNN-GRU model in TensorFlow to separate human micro-movements from environmental noise, at 96.4% validation accuracy.
  • --Local edge processing that broadcasts real-time occupancy to a React dashboard through FastAPI JSON payloads.