Yulin Song

*** **** **., ********, ** +1-917-***-**** ad1tew@r.postjobfree.com Github Linkedin

EDUCATION

New York University New York, NY

M.S. in Computer Science 09/2023 – 05/2025(Expected) Fudan University Shanghai, China

B.S. in Computer Science and Technology 09/2019 – 06/2023 Honor: Third Prize of the Scholarship for Outstanding Graduates at Fudan University (Top 25%) Extracurricular activities: Gold Medal winner of Fudan Tennis Cup; Vice Minister of Academic & Tech Innovation Center SKILLS

• Programming Languages: Java, Python, C/C++, JavaScript, TypeScript, Golang

• Framework: Spring Boot, ReactJS, Django, Flask, Gin, PyTorch, HuggingFace Tools & Models, Gradio, Deepspeed

• Tools & Platforms: Git, Shell, Postman, AWS, Docker, Travis CI, Linux, PostgreSQL, MySQL, ElasticSearch EXPERIENCES

LLM Application Developer Intern Knowledge Works Research Laboratory at Fudan 12/2022 – 06/2023 Peoject: Large Language Models(LLM) Agent for Complex Claims Fact-checking GitHub

• Innovatively researched using of LLMs as Agent for complex fact-checking by leveraging Chain-of-Thought prompt design and In-Context Learning techniques. Employed multi-round dialogues and recurrent self-reflection to construct an efficient full-stack fact-checking platform based on Chatgpt as the core using Python and Gradio framework

• Achieved State-of-the-Art performance on multi-hop datasets, outperforming all leading baseline models by over 5% while enjoying advantages such as excellent interpretability, strong robustness, and low deployment cost

• Wrote a A-Grade thesis with a comprehensive model design and rigorous analysis on the impact of different designs Machine Learning Research and Development Intern University of Maryland, College Park 06/2022 - 10/2022 Mentor: Prof. Jordan Boyd-Graber, CLIP Laboratory, University of Maryland, College Park

• Proposed innovative approaches leveraging perplexity, knowledge and reasoning capabilities of Pre-trained LMs after comprehensive analysis of timeliness to address the limitation of existing timeliness identification methods

• Created a Faiss-indexed database with timestamps by writing Python and Shell scripts to concurrently process and encode over 13 years of Wikipedia data (400GB+) utilized Bert model; Fine-tuned a RAG generator for retrieval

• Developed a retrieval-augmented system in Python with HuggingFace framework to identify the timeliness of questions in knowledge-intensive QA datasets, which outperformed existing leading models by 17% in precision, while maintaining a high rate of recall and offering unparalleled interpretability NLP Research Assistant Knowledge Works Research Laboratory at Fudan 07/2021 – 06/2023 Mentor: Prof. Yanghua Xiao, Fudan University

Engaged in two research projects and was responsible for method implementation and conducting experiments, including:

• Conducted multi-GPU parallel training of LLM using DeepSpeed and Lora frameworks

• Conducted in-depth research into the capabilities of LLMs for step-by-step complex factual reasoning and verification

• Implemented key decoding algorithm of LM and a trie-tree to achieve generative fact-checking with 8% improvement PROJECTS

Full-Stack Project: Smooth Django, ReactJS, AWS, Elastic Beanstalk, Zenhub, Postman GitHub 09/2023 – Present

• Developed a workflow visualization social app using ReactJS and Django, interacting with a SQLite database

• Applied Scrum methodology in development; Utilized Zenhub for project management and Postman for testing

• Implemented RESTful APIs for user registration and management, canvas manipulation and social networking

• Created front-end with React-flow, providing intuitive canvas interaction, dependency check and real-time saving

• Deployed the application to AWS using Elastic Beanstalk, ensuring scalable and efficient application performance.

• Developed test suites achieving 97% backend code coverage measured by Coverall.io integrated with Travis CI Full-Stack Project: ODAP ReactJS, Spring Boot, MySQL, Docker, Postman 04/2023 – 06/2023

• Developed an online data annotation platform using ReactJS and Spring Boot with MySQL database deployed in Docker

• Utilized Spring Security in registration, login, and user session management modules

• Implemented 30+ RESTful APIs for datasets upload, management, preview and annotation with progress tracking

• Designed database schemas with MySQL and interacted with the database using Hibernate. This platform was adopted by many groups of KW Lab and received acclaim for significantly enhancing efficiency Full-Stack Project: Pokemon Battle Golang, Gin, PostgreSQL, Cocos Creator GitHub 04/2021 - 06/2021

• Developed a Pokemon catching and battling game using Gin framework connected with a PostgreSQL database

• Implemented 20+ RESTful APIs handling user registration and login, user profile and Pokemon backpack management, multi-round Pokemon fight support, Pokemon catching and sharing functions

Contact this candidate