Chih-hsin Esa Chen

Led iOS development of a voice-controlled Street View navigation app for blind and low-vision users, validated with Helen Keller Services.

• Shipped four navigation modes with spatialized audio cues and progress tracking
• Engineered on-device ASR with homophone normalization to improve accuracy in outdoor settings
• Coordinated TTS/ASR timing to remove self-echo and reduce announcement latency

Built a Flask-based chatbot querying AWS Bedrock Knowledge Base with advanced filtering and reranking capabilities.

• Engineered admin SPA for knowledge base management with search, metadata facets, and CRUD operations
• Developed data pipeline tooling for FAQ normalization and S3 upload automation
• Implemented per-turn logging for retrieval evaluation and performance monitoring

Full-stack developer on a social cocktail discovery platform with geospatial search and activity feeds.

• Accelerated geospatial query performance by 5× (5s → 1s) by migrating to PostgreSQL + PostGIS
• Engineered core social features: JWT authentication, follow graph, and paginated activity feed
• Resolved AWS EC2 deployment blocker by diagnosing memory issues and packaging custom Conda environment

Built real-time fall detection system with ML-powered health classification for dementia patients.

• Developed real-time fall detection processing IoT data via AWS Lambda with WebSocket alerts
• Created LSTM health classifier in TensorFlow on 5-second windows of HRV/stride data
• Automated daily data pipeline using CloudWatch Events to process S3 sensor files

ML research project fusing acoustic features with transformer models for emotion classification.

• Achieved 85.2% accuracy by fusing MFCC features with BERT on 2K+ utterances
• Observed +2.2 SD pitch shifts in high-arousal emotions via speaker-wise z-normalization
• Applied multimodal deep learning techniques for robust emotion detection