Yunyang Lu

• Designed and implemented zero-shot behavior planning models using bi-simulation techniques, enabling agents to act without demonstrations, reward models, or pre-trained weights. • Developed action-sequence generation algorithms for dynamic object manipulation (e.g., block navigation), improving planning efficiency by 58% in point-maze scenarios. • Containerized workflows with Docker and automated experiment management, ensuring reproducibility and scalability for lab research and downstream applications.

• Engineered a cross-platform mobile app for discovering nearby restaurants and food, collaborating with designers, PMs, and developers to ensure a seamless, engaging user experience. • Designed and constructed backend infrastructure, architecting system via AWS services (EC2, S3, Cognito, CDN), model design, and API design to ensure scalability and reliability. • Implemented and managed 12+ RESTful APIs supporting core functionalities like video uploading and registration using Express.js, Node.js, and MongoDB. • Refactored video API by compression with AWS MediaConvert, reducing the S3 storage and API latency by 30%. • Created 8+ dynamic and responsive front-end components with React Native and TypeScript, including camera integration and reusable components, enhancing development productivity by 20%.

Spearheaded a systematic exploration of the AI system life-cycle, mastering stages from data collection to model evaluation, which provided a holistic view of the machine learning industry. Acquired hands-on experience with time-series data forecasting, web scraping with python for data collection, and statistical analysis for feature engineering, leading to a comprehensive grasp of machine learning models. Concluded reports summarizing machine learning theses, tests on data preprocessing based on dataset statistics, and evaluations of different models and hyperparameters.