Hamna Kaleem

Fine-tuned Llama 3.1 and implemented GPT-2 for a custom GPT application. Conducted training sessions on NLP Text and Audio technologies to improve team proficiency and led internships on audio and text architectures.

Specialized in dataset analysis, data preprocessing, finetuning, and deploying AI architectures with TensorFlow/PyTorch, Flask, Docker, and CUDA technologies. Optimized and customized AI model implementations for noisy, domain-specific, small audio datasets, working on Command Recognition, Sentiment Analysis, Person Identification, and Keyword Analysis.

Worked on ASR implementation and NLP-based mispronunciation detection for Quranic recitation data. Utilized and implemented the Listen Attend and Spell (LAS) architecture for phoneme level speech recognition and investigated effects of explainable AI in ASR systems.

Trained students and professionals in facial recognition, gesture analysis, and gait analysis using PoseNet architecture deployed on PyTorch. Developed machine learning modules using scikit-learn, focusing on temporal and spatial features, and delivered online training sessions for deep learning techniques.

Prepared datasets to analyze linguistic data like word frequency, usage patterns, and associated contexts. Created a front-end interface to display word usage history, text analysis, and contextual examples, using Python libraries to analyze textual data and Flask to connect front-end and back-end services.

Grade: CGPA: 3.7

Activities and societies: Invited Workshop Speaker at the MacViz Computer Vision Workshop, NUST Pakistan (July - September 2021), where I delivered lectures focusing on time series problems, audio signal processing, hybrid NLP CV applications.

Developed a hybrid CNN-RNN architecture for emotion detection using gait analysis, simultaneously analyzing spatial and temporal features. Implemented the model with TensorFlow/Keras, achieving 86% classification accuracy. This work contributes to emotion detection via gait analysis using IMU sensor data.

Grade: CGPA: 3.54

Activities and societies: Silver Medalist for outstanding academic performance and recipient of Full Merit Scholarship throughout the duration of the degree.

Developed a time-series forecasting model to predict Consumer Price Index (CPI) trends using advanced techniques like XGBoost and RandomForest. Enhanced predictions by integrating an LSTM-based RNN and fully connected neural networks. Achieved a 20% reduction in forecasting errors compared to traditional methods like ARIMA.