Kate — Android AI Assistant (Personal Project)
Trained a custom NLP intent-classification model to power an on-device Android assistant with real-time intent recognition. The work involved preparing text inputs for modeling through tokenization, sequence padding, and supervised training with sparse categorical cross-entropy loss. The resulting model was converted to TensorFlow Lite (TFLite) format for efficient deployment on mobile devices. • Built an intent-classification pipeline from raw text to model-ready sequences • Trained and validated the intent model using TensorFlow/Keras in Google Colab • Converted the trained model into an on-device TFLite model for inference • Integrated model outputs into an intent-routing and event-dispatching flow in the app