AI Research · Biomedical Signal Processing · Brain–Computer Interfaces
Computer Engineering student at Menoufia National University
GPA: 3.97/4.00 · Class Rank: 1st
Research prototype for classifying normal and abnormal heart-sound recordings, with a WAV-upload demo, model artifacts, and evaluation scripts. Reports recording-level benchmark results on the PhysioNet/CinC 2016 dataset; it is not clinically validated or intended for diagnosis.
Reproducible experiments on model capacity, interpolation, regularization, and generalization, including a nested-cross-validation diabetes benchmark.
University of Louisville — Louisville, Kentucky, USA
Summer Research Intern · July–September 2026
- Developed an EEG window-prioritization pipeline processing 616 hours of continuous recordings.
- Reduced the data requiring detailed analysis by 92.5%, retaining 79.6% seizure-event sensitivity.
- Built a feature-extraction and XGBoost classification workflow.
- Built a complete heart-sound classification pipeline using 3,240 recordings.
- Engineered 301 signal features using MFCCs, spectral statistics, and wavelet transforms.
- Achieved 93.6% accuracy and 0.975 AUROC on a held-out test set.
- Adapted the MIRepNet EEG model to 15-channel OpenBCI recordings.
- Implemented signal preprocessing, resampling, Euclidean Alignment, spatial interpolation, and transfer learning.
- Achieved approximately 79% macro accuracy under strict held-out-participant evaluation.
- Developed an SSVEP-based directional-control experiment.
- Conducted a structured review of brain–computer interfaces for robotic systems.
- Organized evidence from 1,838 initially identified records into a structured research taxonomy.
| Area | Technologies & Methods |
|---|---|
| Machine Learning | PyTorch, scikit-learn, XGBoost, LightGBM, transfer learning, ensemble learning |
| Signal Processing | EEG, OpenBCI, MFCCs, spectral analysis, wavelet transforms, feature engineering |
| Model Evaluation | AUROC, sensitivity, specificity, cross-participant validation, uncertainty-aware classification |
| AI Engineering | Flask, Docker, CI/CD, inference pipelines, reproducible experimentation |
- Mitigating Data Bias in AI-Assisted Neurosurgery
- Mitigating Double Descent in High-Dimensional Models
- SHARP: Wi-Fi Sensing and Recognition Methodologies