Hi, I'm Sophie Shi

• Building a multimodal system that integrates video, images, and brain signals.
• Developing a vision language model (VLM) pipeline for automated behavior detection in clinical video data.
• Performed supervised fine-tuning (SFT) using Sapiens (Meta) and MMPose (OpenMMLab).

• Built an LLM tool using Qwen-3 to extract insights from arXiv papers via single paper summarization or multi-document RAG-augmented summarization.
• Integrated semantic chunking, BAAI/bge-small-en embeddings, FAISS search, and structured prompts for targeted summarization.

• Implemented a multi-camera algorithm to extract 3D pose landmarks of freely moving mice.
• Developed an automated pipeline for quantitative behavior profiling beyond subjective manual assessments.

• Set up a multi-camera motion capture system for specific diagnostic tasks and daily activity recording.
• Performed hand tracking analysis (finger tapping) and gait analysis for Parkinson’s disease behavior phenotyping.

Developed joint compression-classification network using ResNet architecture for EEG seizure prediction. Implemented adaptive reconstruction module with under variable compression ratios for enhanced prediction accuracy.

Built virtual machine-based RESTful API server and client with persistent database integration. Developed client GUI and monitor-end GUI applications for medical record and real-time patient monitoring.

Designed multi-view iOS/WatchOS mobile application for deadline management using SwiftUI and UIKit. Integrated cross-device real-time synchronization via iCloud and Firebase, with Sakai API integration for automatic course assignment deadlines.