Muhammad H.

Fine-tuned and trained Large Language Models (LLMs) on open-source codebases to enhance code understanding, generation accuracy, and domain adaptation. Designed and implemented comprehensive AI solutions that encompassed data preprocessing pipelines, model training workflows, evaluation frameworks, and deployment strategies. Led the implementation of MLOps for a technical course project, developing automation scripts for model training, validation, CI/CD integration, and deployment workflows. Built reproducible ML pipelines utilizing Python, Docker, and cloud tooling to support scalable model deployment. Developed and refined prompt engineering strategies to improve LLM performance for tasks related to code generation and technical documentation. Conducted rigorous model evaluation with quantitative metrics and error analysis to iteratively enhance model accuracy and reliability. Wrote clean, maintainable, and production-grade code, adhering to best practices for scalability. Collaborated effectively with clients and cross-functional stakeholders to transform business requirements into technical AI solutions. Engaged with frameworks and tools including PyTorch, Hugging Face Transformers, Git, Docker, MLflow, and various cloud environments.

Conducted independent research focused on adversarial attacks against machine learning and computer vision systems, enhancing deep Python expertise while implementing AI/ML security solutions. Successfully executed HopSkipJump attacks on distributed vision systems to uncover significant vulnerabilities in autonomous systems, achieving a 100% success rate on VPR and 70% on 3D Reconstruction through containerized environments. Developed reproducible experiments and cloud-based prototypes to identify weaknesses in AI systems while employing workflow orchestration techniques for systematic testing. Regularly shared research findings via internal documentation and presentations, emphasizing transparency and collaboration with distributed teams. Created automated Python scripts to benchmark ML model performance under adversarial conditions and validated system robustness through Linux-based testing environments. Produced detailed technical reports and visualizations of experimental outcomes, providing a foundation for future engineering implementations while showcasing strong data analytics skills.

Improved TeekSho's online presence through web development, implementing cloud solutions contributing to the startup's revenue growth. Worked collaboratively with cross-functional teams using agile methodologies to redesign and enhance the website architecture, ensuring a modern and responsive interface while significantly reducing response time through distributed systems concepts. Developed and integrated sophisticated ETL processes and features for data analytics utilizing containerization technologies, actively participating in open-source development practices to adapt to evolving user requirements and improve team efficiency.