research Vision-Language Coach for Ingestive Behaviors Fine-tuning Qwen2.5-VL with LoRA to jointly classify eating-behavior quality and generate clinician-style coaching feedback (NIH R01 — DIBS). Tripolar EEG — Three-Way Electrolyte Comparison Methods/pilot study comparing Paste, Gel, and saline-soaked Felt TCRE configurations on resting-state and visual-evoked-potential recordings. PRIME — Closed-Loop LLM-Based Cobot Computer-vision perception module for a closed-loop human-robot teaming cobot driven by LLM-based autonomy allocation. EEG + Motion Capture for Movement Decoding Hardware synchronization of high-density g.tec EEG with optical motion capture for decoding grasping and reaching movements. Automated Neuromuscular Assessment with Robotics + CV Marker-free 3D hand pose estimation and closed-loop feedback in unsupervised rehabilitation sessions, mediated by Socially Assistive Robots. code YCB Object Segmentation with YOLOv26 Fine-tuned YOLOv26n-seg for segmenting YCB benchmark objects in images and video. Supports a perception module for robotic manipulation. EEG Motor-Intention Classification (FBCSP) Filter Bank Common Spatial Patterns pipeline for decoding motor intention from EEG; 84.7% mean accuracy. 3D Hand Pose Estimation from RGB CNN with transfer learning to recover 3D hand keypoints from monocular RGB images. Motor Unit Pair Analysis (BSS Decomposition) Post-processing analysis of motor unit pair behavior following blind source separation decomposition of high-density EMG.