About Me
AI researcher and engineer studying how AI systems, networks, and human societies shape one another.
Who I Am
I'm Chengyu Li, an M.S. student in Artificial Intelligence at Northeastern University. My background spans both AI research and industry: I have built and deployed LLM systems in production, and I now use computational methods to study how information, influence, and social context move through human systems.
My research sits at the intersection of AI, network science, and computational social science. I study large-scale science-to-policy networks, examine how social context is represented inside language models, and am interested in developing rigorous ways to use AI agents as models of social systems. Across these problems, I care about the same question: how do AIs and human societies shape each other's behaviors and structures?
Research & Industry Experience
Harvard University
Research Intern, Health System Innovation Lab | Sep 2025 - Present | Boston, MA
- Study how oncology research moves into policy through a global network covering 204 countries and approximately 4.7 million fractional publications.
- Build directed, weighted science-to-policy networks and use null models, community detection, distributional inference to separate structural patterns from scale effects.
- Proposed Expected Absolute Cumulative Citations (EACC), a Kaplan–Meier-inspired metric incorporating right-censoring logic for long-horizon policy-citation estimation to evaluate the temporal effect of differernt types of research on different policies.
- Examine inequalities in whose research reaches governments, intergovernmental organizations, and other policy actors, and build interactive D3.js visualizations of global knowledge flows.
Northeastern University
Research Project, Actionable Interpretability | Feb 2026 - May 2026 | Boston, MA
- Studied how U.S. neighborhood names shift policy framing in Gemma-2-9B-IT, connecting model behavior to social context rather than treating prompts as context-free inputs.
- Built a behavioral, representational, and causal interpretability pipeline using linear probing, activation patching, and contrastive steering.
- Found that instruction tuning creates a systematic place-name framing effect and that strong probe signals do not necessarily imply causal control over model behavior.
KPMG China
Gen AI R&D Intern | May 2025 - Aug 2025 | Nanjing, China
- Participated in 5+ enterprise-level digital transformation projects using LLMs and generative AI.
- Implemented RAG systems with LangChain, RagFlow, FastAPI, and Milvus.
- Improved retrieval quality with semantic chunking and hierarchical indexing, increasing Top-20 accuracy by 20% while reducing latency by 50%.
- Built containerized MCP servers and LLM-agent workflows with FastMCP and Docker.
- Contributed to KPMG's next-generation AI platform integrating Qwen and DeepSeek with vLLM.
Northeastern University
Research Assistant, GUI Agent Safety | Mar 2025 - Sep 2025 | Boston, MA
- Studied how GUI agents degrade under prompt-injection and visual-distraction perturbations, treating the agent-environment interface as the unit of analysis.
- Constructed large-scale multi-turn benchmarks and implemented ViT-based distraction analysis to compare robustness across deployment settings.
Conlight Medical
LLM Engineer Intern | Dec 2023 - May 2024 | Shanghai, China
- Developed an end-to-end LLM chatbot for mental health using Python, FastAPI, and MongoDB.
- Led dataset construction and preprocessing for 50K+ anonymized clinical records.
- Implemented SFT and RLHF workflows with Llama Factory to improve model alignment.
- Deployed B2B and B2C web demos with vLLM for low-latency inference.
Education
Northeastern University
M.S. in Artificial Intelligence | Sep 2024 - Dec 2026 | Boston, MA
GPA: 3.89/4.0
Coursework: Advanced Machine Learning, Deep Learning, Algorithms, Large-Scale Data Processing, LLM-Based Dialogue Agents, Actionable Interpretability, Abolition Technology
Nanjing Normal University
B.Eng. in Computer Science | Sep 2020 - Jun 2024 | Nanjing, China
GPA: 3.84/4.0
Coursework: Linear Algebra, Probability & Statistics, Calculus, Discrete Mathematics, Distributed Systems, Object-Oriented Programming
Publications & Manuscripts
- DFEN: Dual Feature Equalization Network for Medical Image Segmentation[Link]
- Swin-TransUper: Swin Transformer-Based UperNet for Medical Image Segmentation[Link]
- Global Oncology Knowledge Flow Networks: Quantifying Science-to-Policy Translation — manuscript[Link]
- Auditing the “Carceral Logic”: How Neighborhood Names Shape Policy Framing in Large Language Models — manuscript[Link]
- Global Landscape of Sepsis Research Funding, Publications, Clinical Trials, Patents, and International Collaboration Networks, 2000-2025 — manuscript
Selected Engineering Projects
Haper-IO | LLM-Based Email Agent System
- Designed an LLM-powered email agent using the GPT-4 API, Redis caching, and PostgreSQL.
- Implemented long-term memory management with a memory bank and forgetting-curve mechanism.
- Reduced redundant storage by 30% and improved long-term contextual accuracy by 20%.
National Innovative Entrepreneurial Project | Computer Vision System
- Led a team of four developing a real-time vehicle detection and tracking system with YOLOv5 and OpenCV.
- Built lane-recognition and tracking components and delivered the system in a production environment.
Teaching
Northeastern University
Teaching Assistant, CS 6140 Machine Learning | Jan 2026 - May 2026 | Boston, MA
Led recitations and weekly office hours, designed problem sets, and graded graduate course projects.
Technical Skills
Network & Computational Social Science
NetworkX, igraph, Gephi, Louvain, Leiden, Infomap, null models, survival analysis, D3.js, Three.js
ML & Interpretability
PyTorch, Hugging Face Transformers, TransformerLens, vLLM, Llama Factory, linear probing, activation patching, contrastive activation steering
LLM Systems
RAG, LangChain, RagFlow, FastMCP, FastAPI, Milvus, Redis, model fine-tuning, agent workflows
Programming & Infrastructure
Python, C++, JavaScript/TypeScript, SQL, Git, Linux, Docker, AWS, HPC clusters
Certifications & Honors
- AWS Certified Solutions Architect - Associate
- MIT Sloan Executive Certificate - Digital Business
- National Third Prize, Chinese Collegiate Computing Competition (2024)
What's Next
I'm applying to PhD programs where I can study AI as part of a broader social system rather than only as a model in isolation. I'm especially interested in research that combines AI, network science, and computational social science to understand collective behavior, information and influence flows, and the feedback loops between AI systems and human societies.