Satvika S

Predicted HIV-1 resistance to 24 antiretroviral drugs by combining ESM-2 protein language model embeddings with GraphSAGE to capture co-occurring mutation interactions. Built a full pipeline on Stanford HIVDB data (sequence curation, subtype/polymorphism correction, confidence-weighted labeling); achieved high AUROC.

Integrated SHAP-based explainability and epistasis analysis to surface clinically relevant mutation combinations driving resistance.

Built a writer identification system using ResNet-18 to extract 512-dim handwriting embeddings from camera-captured documents, requiring no specialized biometric hardware. Evaluated 10 classifiers plus stacking ensemble across original, PCA, and SFS feature sets; stacking on full embeddings achieved 97% test accuracy.

Applied SHAP and LIME for feature-level explainability, making the system interpretable for real-world deployment.

Designed a deadlock-free token passing solution with priority scheduling and thread synchronization, guaranteeing bounded waiting and starvation freedom under concurrent execution.

Member of one of India's top competitive cybersecurity teams.