ATIFA NAWAZ

• Built an AI research assistant capable of processing PDFs, DOCX files, and online paper links to provide context-rich answers.
• Integrated LangChain, FAISS, and LLaMA 3 for retrieval-augmented generation and citation-aware responses.
• Implemented document parsing, embedding, and inline citation injection for accurate and verifiable results.
• Developed a robust question-answering pipeline with semantic search over document embeddings.

• Developed an AI-powered chest X-ray diagnostic web application using CNNs for classification.
• Implemented explainable AI visualizations using GradCAM++ to highlight regions of interest for model predictions.
• Deployed a Streamlit interface enabling real-time image upload, prediction, and interactive visualization for research and educational purposes.

• Developed an Artificial Neural Network (ANN) to predict loan defaults using historical loan data, handling imbalanced datasets with SMOTE and scaling numeric features.
• Designed the ANN architecture with dense layers, ReLU activations, batch normalization, dropout, and a sigmoid output layer for binary classification.
• Evaluated model performance using Accuracy, Precision, Recall, F1-Score, and AUC, achieving high recall (0.8927) to minimize risk and visualized results with ROC curves and training history plots.

Performed end-to-end data cleaning, preprocessing, and exploratory data analysis using Pandas and NumPy on real-world datasets. Built and evaluated regression models with Scikit-Learn to support data-driven predictions and insights.

Designed and trained ANN and CNN models using TensorFlow/Keras for supervised classification and regression tasks. Analyzed model performance using loss curves, accuracy metrics, and validation strategies.

Applied instruction tuning and reinforcement learning methods (RLHF, PPO, DPO) to improve LLM performance using Hugging Face. Fine-tuned transformer-based LLMs and evaluated results on real-world natural language processing tasks.