Hong Quan TRAN

During my time at HephIA, a startup specializing in innovative machine learning solutions, I’ve had the privilege of working in a dynamic and versatile role that allowed me to experiment across multiple responsibilities. 2023 - 2024: • Product: Lawyer assistant using modern NLP to streamline legal workflows (Python, CI/CD, Docker, AWS) • Fine-tuned and deployed NLP models for legal NER tasks (e.g., case numbers, law references), anonymizing private information. The model had a more than 90% true positive rate. (Hugging Face, Transformers, PyTorch, AWS) • Designed and developed AI Agent Systems to automate repetitive legal tasks (document summarization, automated legal drafting, ... ), significantly reducing task time by 20% and boosting productivity. (Langchain, LangGraph, LLM, Transformer) • Participated in designing an OCR pipeline for data collection and preprocessing using Tesseract and EasyOCR. • Built and integrated RAG pipelines into production systems to support internal search, collaborating with cross-functional teams to deliver robust, client-customized solutions. (LangGraph, LangChain, MongoDB, ...) • Monitored model performance by implementing MLOps pipelines, ensuring reproducibility and data integrity across experiments and deployments. (MLflow, AWS, Docker) 2022 - 2023: • Product: Scalable System focusing on running Unsupervised Learning algorithms. Integrated Unsupervised Learning algorithms modules into the code base (Python, Scala, Spark, GitHub) • Analyzed large-scale client datasets to uncover data outliers and common patterns. (Scikit-learn, Pandas, ...) • Participated in code testing cycles, maintained up-to-date documentation with Sphinx, provided functional feedback, and helped improve user experience through peer reviews and clear documentation. (GitHub, CI/CD)

During my internship at HephIA, under the supervision of Professor Mustapha Lebbah (Paris-Saclay University - UVSQ-Versailles), I was tasked with: • Designed custom visualizations of airplane engine data derived from the Self-Organizing Maps (SOM) unsupervised learning algorithm. (Bokeh, Matplotlib, and Seaborn) • Integrated custom visualizations into the analytical product developed by HephIA. (PySpark, Python, CI/CD)