Hemalathaa Arumugam Elumalai

Master's student in Artificial Intelligence and Cybersecurity with a strong foundation in computer science, data engineering, machine learning, and deep learning. Experienced in data analysis, AI-driven projects (NLP, computer vision, generative AI), and hands-on cybersecurity skills such as penetration testing and credential harvesting. A proactive learner with proven leadership and collaboration abilities, eager to leverage my expertise in both AI and cybersecurity to drive innovative solutions in a dynamic environment.

• Network Security-Email Phishing Detection using MLOps: Developed and deployed a phishing URL detection system utilizing MLOps practices. Led end-to-end project implementation, from model training to deployment, using tools like Python, Docker, Git, FastAPI, and MongoDB. Automated workflows with CI/CD pipelines and Airflow, ensuring efficient model updates and scalability. Managed infrastructure with Terraform, and used AWS S3 for data storage, integrating Starlette for seamless API responses. The solution effectively detects phishing URLs in real-time to enhance network security.

      https://github.com/hemalathaa90/Network-Security.git

• MCQ Generator using OpenAI: Developed an AI-driven multiple-choice question generator using OpenAI’s GPT API for automatic question and answer creation from text and PDF files (extracted with PyPDF2). Implemented LangChain to manage multi-step NLP workflows and built an interactive user interface with Streamlit. Ensured secure handling of API keys and environment variables using Python-dotenv, https://github.com/hemalathaa90/mcqgenerator
• Huggingface with Langchain -Text Generation Project: Built and fine-tuned text generation models using Huggingface, alongside a Python library called histLang to generate Seq2Seq style and encoder-decoder type of language model. We implemented decoder-only architectures tuned as a balance between performance and language coherence in the generated outputs.
• Skin Cancer Detection: Initiated a rigorous data cleaning and preprocessing phase, addressed missing values, duplicates, and outliers, followed by dataset balancing through undersampling and oversampling techniques, leading to the successful training of models on MobileNet, VGG19, and ResNet50 architectures, achieving an impressive average accuracy of 98%, precision of 0.9, recall of 0.8, and an F1 score of 0.9, thereby demonstrating the critical role of meticulous data preparation in significantly improving model performance and accuracy., https://www.kaggle.com/code/hemalathaaae/mobilenet-skin-melanoma
• Hand Object Detection: Developed a model to protect against heavy machinery accidents by detecting both open and closed hand positions through a video stream using TensorFlow Object Detection, incorporating confidence level indications and alarm sounds to ensure hands remain in a safe zone., https://github.com/hemalathaa90/Hand-Detection-Using-TFOD