I build reproducible applied AI systems across computer vision, medical AI, remote sensing, vision-language workflows, sustainability applications, and environmental machine learning.
My current portfolio is centered on five finalized Python-first projects. Each repository emphasizes clear documentation, reproducible structure, evidence tracking, limitation awareness, and responsible project framing.
Portfolio website: mohamad-sabbagh-ai-research-portfolio--za9699185.replit.app
| Order | Project | Focus | Repository |
|---|---|---|---|
| 1 | ECG Image-to-Signal Reconstruction | Medical AI, computer vision, signal reconstruction | medical-ecg-image-to-signal-reconstruction-pipeline |
| 2 | Satellite Land Classification with CNN/CNN-ViT | Remote sensing and comparative vision architectures | satellite-land-classification-cnn-vit |
| 3 | Waste Classification using VGG16 Transfer Learning | Sustainability-focused image classification and model release | waste-classification-transfer-learning |
| 4 | Aircraft Damage Classification + BLIP Reports | Inspection-support computer vision and vision-language workflow | aircraft_damage_vgg16_blip |
| 5 | Rainfall Prediction in Australia | Tabular ML, metric provenance, and environmental prediction | rainfall-prediction-classifier |
- Medical AI and signal-aware computer vision — ECG image-to-signal reconstruction with synthetic benchmarking, QC checks, failure-mode analysis, and pipeline compatibility tooling.
- Remote sensing and visual classification — CNN/CNN-ViT experimentation for agricultural vs non-agricultural land classification, with metric provenance and limitation tracking.
- Applied inspection AI — aircraft damage classification combined with BLIP-based caption/report generation, framed as inspection-support rather than certified maintenance tooling.
- Sustainability-focused ML — VGG16 waste classification with a bundled checkpoint, direct inference script, model-release documentation, and responsible use boundaries.
- Environmental tabular ML — rainfall prediction with classical ML models, leakage/split-risk awareness, temporal validation protocol, and calibration/interpretability planning.
- Python-first ML repository design
- Computer vision and transfer learning
- TensorFlow/Keras and PyTorch workflows
- Classical ML with scikit-learn and XGBoost
- Research evidence packs, metric provenance, and reproducibility checklists
- Responsible project framing: limitations, safe claims, and future-work separation
A detailed research portfolio command center is available here:
AI Research Portfolio Command Center
It organizes the five projects, safe claims, CV wording, professor reading paths, and final packaging steps.
These repositories are research and portfolio projects. They do not claim clinical validation, production deployment, certified inspection readiness, state-of-the-art status, or operational forecasting service status unless explicitly supported in the relevant repository documentation.