AI Researcher & Engineer · MASc Electrical and Computer Engineering @ Ontario Tech University
Efficient deep networks · Synthetic media & deepfakes · Trustworthy AI
- 🎓 MASc in Electrical and Computer Engineering at Ontario Tech University, in the IoT Research Lab with Prof. Khalid Elgazzar
- 🔬 I do AI research across model efficiency, synthetic media, and trustworthy AI
- 🛠️ I ship real systems too: LLM agents, RAG pipelines, APIs, and apps
- 🎥 I make short videos on how AI works on The Mo Lab
- 🩺 Background in Biomedical Engineering: imaging, signal processing, and hardware
| Area | Work |
|---|---|
| ⚡ Efficient deep networks | LAGD-Net: Differentiable Block Gating for Depth Pruning of Residual Networks (in progress). Learned gates drop whole residual blocks during training: 14 of 33 blocks removed on CIFAR-10 ResNet-101, 1.63× faster at −0.17 points. |
| Block-selection benchmark for depth pruning (in progress). Which blocks to remove, and does it matter after fine-tuning? 8 architectures, 3 datasets, 740 runs. | |
| 🎭 Synthetic media & deepfakes | MOZAIC: audio-video deepfake benchmark that tests what is fake (face, voice, or both). 137K clips, 1,000 identities, with Vector Institute. Code |
| 🛡️ Trustworthy AI | When Does Test-Time Augmentation Help Calibration? VMV 2026. Paper · Code |
| 🩺 Medical AI | AI-based tool for early detection of Alzheimer's disease. Heliyon 2024. 98.17% accuracy on ADNI. Paper · Code · Demo |
| 🔐 Medical data security | Smart Encryption: A Novel Hybrid Approach to Selectively Encrypt Medical Images. IEEE NoF 2025. AES-256 on the heart ROI, ChaCha20 on the rest, 60–70% faster than full AES. Paper · Code |
| Project | What it does | Stack |
|---|---|---|
| BrainAware: Early Alzheimer's Detection | Classifies brain MRI into CN / MCI / AD (98.17% acc) + web app with 2D/3D viewer and patient history. Heliyon 2024 · Live demo | TensorFlow, Django, React |
| MOZAIC | Audio-video deepfake benchmark: 137K clips, 4-way labels (real/fake face × real/fake voice), 12 MLLMs + 5 detectors evaluated. With Vector Institute | Python, FaceFusion, Wan2.2, Seed-VC |
| When Does TTA Help Calibration? | Test-time augmentation and calibration on 6 MedMNIST datasets, 3 backbones, 9 fusion methods. Entropy weighting cuts ECE by up to 41%. VMV 2026 | PyTorch, MedMNIST |
| BioMed Lab | 15 of my biomedical desktop apps rebuilt as browser tools: CT reconstruction, MRI viewer + 3D, ECG aliasing, Z-transform filter design, sequence alignment. Try it · Code | React, TypeScript, Web Workers |
| Auto-Med-Agent | Generative AI research assistant: RAG over medical papers, PubMed and clinical-trial search tools, summaries, and question generation | Transformers, FAISS, Streamlit |
| AI Resume Screener | Parses resumes (local or Groq LLM), scores them against a job with spaCy + Sentence-BERT, and ranks applicants | FastAPI, PostgreSQL, Next.js |
More projects: imaging, signals, ML, and software
Most of the imaging, signal and bio projects below now run in the browser in BioMed Lab.
| Area | Projects |
|---|---|
| Medical imaging | MRI Planes Viewer · Medical Image Rendering (VTK) · Exudate Detection in Fundus Images · Image Processing Tools from Scratch · Cervical Spine Fracture (RSNA 2022) · Trust in AI Retinal Screening (UI prototype) |
| Signals and hardware | Angry Birds Controlled by EMG (Arduino) · Digital Signal Processing Projects |
| ML and DL | COVID-19 Prediction · Machine Learning Models · Deep Learning Practice (AE, VAE, GANs) |
| Bioinformatics | Sequence Alignment App (PyQt) |
| Software | Algorithms Toolbox · Group Funding Website · Secure Coding · Smart Hospital Simulation |