Undergraduate research on multi-model generative pipelines and HIPAA-compliant medical NLP — under advisors Dr. Chengwei Lei and Dr. Kanwal Kaur at CSU Bakersfield.
Multi-model video-generation pipeline combining YOLO11 detection, BLIP2 captioning, and AnimateDiff diffusion. Uses confidence-weighted keyword fusion and a temporal-chaining scoring function for cross-frame visual consistency.
HIPAA-compliant bidirectional medical translation system for real-time doctor–patient communication. Whisper speech-to-text feeds MarianMT and NLLB-200 translation, then a TTS layer renders the response. Designed for clinical deployment with audit logging and on-device inference where feasible.
Two-stage detection pipeline for crop-disease severity — YOLOv11x for lesion localization, HSV-space analysis for severity grading. Training used cosine learning-rate scheduling with warmup and standard augmentation. Built a per-class evaluation suite (ROC curves, confusion matrices) and an automated overfitting-diagnostic dashboard. Planned ablation study addresses full-image bounding-box limitations identified in mentor review.