CV

Jiang Shutong (江姝潼)

You can download the full CV here.

Tsinghua University
- M.S. in Software Engineering, School of Software
- Sept. 2023 - June 2026 (Expected)
- Supervised by Prof. Yingbo Liu at the Tsinghua Thulab
Beijing University of Posts and Telecommunications (BUPT)
- B.E. in Network Engineering, School of Computer Science
- Sept. 2019 - June 2023
- Ranked 1st in major, recommended for admission to Tsinghua University

DeepPlanning: Benchmarking Long-Horizon Agentic Planning with Verifiable Constraints Shutong Jiang, et al.
arXiv preprint, 2026. [Link]
Qwen-3 Technical Report An Yang, Anfeng Li, …, Shutong Jiang, …, et al.
arXiv preprint, 2025. [Link]
Qwen3-VL Technical Report Shuai Bai, Yuxuan Cai, …, Shutong Jiang, …, et al.
arXiv preprint, 2025. [Link]

Duties: Algorithm Intern (Agent Group)

MCP Integration & Paradigm Optimization: Designed a “Planning-Action-Reflection” framework for Qwen3 Agents, enhancing model reasoning and task decomposition.
Qwen-VL Agent System: Constructed a three-stage data pipeline for multimodal tool-calling, introducing “Zoom-in & Search” patterns for complex visual reasoning.
Agent Pre-training: Developed a large-scale data generation pipeline for tool-use, producing over 2.38M high-quality trajectories (9.6B tokens) to internalize agentic capabilities.

Duties: Algorithm Intern (Alignment & RL Group)

Novel Writing Agent: Built a Multi-Agent framework utilizing Role-Playing paradigms to automate long-form fiction creation.
Dynamic Memory Mechanism: Implemented an LLM-driven memory system with sliding window updates to ensure character and plot consistency in large-scale narratives.

Duties: Algorithm Intern (Edge Machine Learning Group)

Edge Copilot Optimization: Fine-tuned 8B-scale open-source models for automatic browser tab grouping, achieving a 78% user click-acceptance rate.
Fine-grained Alignment: Utilized DPO and divide-and-conquer strategies to align model outputs with user preferences for balanced grouping granularity.

Programming: Proficient in Python; experienced with PyTorch and data science libraries (Numpy, Pandas, Matplotlib).
Professional: Deep understanding of LLM Agents, RAG, and fine-tuning techniques; skilled in Llama-Factory and Megatron frameworks.
Language: Excellent English proficiency (CET-6: 586, TOEFL: 97), capable of writing professional technical documentation.