AI Researcher · Software Engineer · XAI & Quantum ML
Engineering explainable AI systems where accuracy alone is not enough.
I am a Master's student in Artificial Intelligence, working at the intersection of Explainable AI and Quantum Machine Learning. My focus is on breaking down the black box of deep learning models for high-stakes domains — particularly clinical diagnostics — where interpretability is not a nice-to-have but a requirement for adoption.
Alongside research, I build production-grade systems: end-to-end ML pipelines, mobile applications in Flutter, and deployed inference services.
Explainable AI for healthcare. Applying feature attribution methods (SHAP, LIME, Integrated Gradients, Grad-CAM) to make medical diagnostics interpretable for clinicians, not just data scientists.
Quantum Machine Learning. Building hybrid quantum-classical neural networks and studying whether the added complexity can be justified through interpretability analysis.
Trustworthy decision frameworks. Combining deep learning architectures (GNNs, Transformers) with rule-based interpretable layers like Fuzzy Logic.
| Domain | Tools |
|---|---|
| Explainable AI | SHAP · LIME · Integrated Gradients · Grad-CAM · EigenCAM · Fuzzy Logic |
| Deep Learning | PyTorch · TensorFlow · Keras · Scikit-Learn · GNNs · LSTMs |
| Computer Vision | YOLOv8 · ONNX Runtime · OpenCV · Ultralytics |
| Quantum ML | PennyLane · Hybrid QCNNs |
| Data Science | Python · Pandas · NumPy · Matplotlib · Seaborn |
| Mobile | Flutter · Dart · Firebase · Supabase · REST APIs |
| Deployment | Hugging Face Spaces · Gradio · Docker · Git |
YOLOv8s · EigenCAM · ONNX · Gradio
End-to-end medical imaging system that detects 8 classes of chest X-ray abnormalities and explains each detection with an EigenCAM heatmap. Trained on 4,380 VinDr-CXR scans using Weighted Boxes Fusion of three radiologists' annotations. Achieves mAP@0.5 = 0.331 with strong per-class performance on anatomical features (Aortic enlargement AP 0.878, Cardiomegaly AP 0.864).
PennyLane · TensorFlow · SHAP · Integrated Gradients
Hybrid quantum-classical neural network for liver disease prediction, integrating multi-method explainability analysis. Associated with a research preprint on trustworthy quantum ML for clinical diagnostics.
LSTMs · GNNs · Transformers · Fuzzy Logic
Multimodal fake news detection pipeline combining sequence models, graph networks, and transformers, with a rule-based fuzzy logic layer for interpretable trust scoring.
- CNN-Based Early Autism Detection — Facial image analysis with Xception and VGG16 architectures
- XAI Diabetes Prediction — Stacking ensemble with LIME-based feature-level explanations
- High-Resolution Face Restoration — U-Net Autoencoders, GANs, and Hybrid Context Encoders
- Smart Helmet App — IoT-integrated Flutter app for accident detection
- InfoKlub App — Encrypted personal data ecosystem with AI-powered CV generation
Towards Trustworthy Quantum Machine Learning in Clinical Diagnostics A Multi-Method Explainability Study of a Hybrid Quantum-Classical Neural Network for Liver Disease Detection
ResearchGate preprint: 10.13140/RG.2.2.11569.34403
- Refining quantum-classical explainability frameworks and human-interpretable validation methods
- Building deployment pipelines that bring research models to production
- Open to research collaborations in XAI, medical imaging, or trustworthy AI
Ask me about: why your model acts like a black box, and how to add transparency layers that clinicians actually trust.
"AI systems used in healthcare, security, and public decision-making must prioritize interpretability, accountability, and human trust — not just predictive accuracy."