AI / ML Intern
BISAG-N, Ministry of Electronics & IT (MeitY)
— Government of India
- Built a full-stack medical imaging platform (React/Vite + Flask) for binary pneumonia detection and 4-class brain tumour classification (Glioma, Meningioma, Pituitary, Normal); MobileNetV2 reached 90.06% accuracy and AUC-ROC of 0.97 on a 624-image chest X-ray test set.
- Benchmarked 5 architectures for pneumonia and 3 for brain tumour — implemented an ensemble majority-voting mechanism with MobileNet tie-breaker fallback, resolving classification across 8 models simultaneously.
- Integrated a Grad-CAM explainability pipeline surfacing pathological heatmap overlays on X-rays and MRIs; added OOD detection to reject non-medical inputs and flag low-confidence predictions (<85%) for radiologist review.