Mazhar Akhtar Ansari

No project description available.

No project description available.

Custom PyTorch C++/CUDA operator vulnerability benchmark with ASAN-based memory safety testing and numerical verification.

Context-aware RAG chatbot using LLMs, LangChain, and FAISS for accurate, grounded responses. Implements document retrieval, semantic search, and multi-turn conversational memory to reduce hallucinations and improve reliability. Features an end-to-end pipeline with a Streamlit-based interactive UI.

Multi-class skin cancer classification pipeline using Vision Transformers (ViT) and CNN architectures like EfficientNet and DenseNet. Includes extensive experimentation, hyperparameter tuning, and data augmentation on imbalanced datasets, achieving high accuracy and robust performance.

Hybrid dual-path deepfake detection system combining CNN-based pixel features and ASCII-based structural representations. Implements spatio-temporal modeling using CNN + BiLSTM for robust video classification. Achieves ~95% accuracy with optimized training, feature fusion, and improved generalization.