Back to Projects
Offline Facial Recognition System
Computer VisionMobileOn-Device ML

Offline Facial Recognition System

2024 · Live Project
Most facial recognition systems phone home to a server. I wanted to build one that runs completely offline — useful in places with no reliable connectivity. Built a Flutter app using MediaPipe and TensorFlow Lite for on-device inference, with embedding-based matching for 100+ stored faces. Achieved ~95% recognition accuracy. The constraint of 'no cloud' turned out to be the interesting design problem.

OUTCOME

~95% accuracy, 100+ user embeddings, zero cloud dependency

TECH STACK

FlutterMediaPipeTensorFlow LitePython

More Projects

Cataract Detection System
01
Computer VisionDeep LearningPython

Cataract Detection System

Real-time cataract detection from eye images. Built the full pipeline — model, inference API, and serving layer. 98.1% accuracy.

98.1% detection accuracy on real eye image datasets
Audio Captioning System
03
Deep LearningAudio MLNLP

Audio Captioning System

Feed it audio. It tells you what's happening. CNN-LSTM and Transformer architectures on extracted audio features.

Contextual captions generated from raw audio without transcription