CV

Education

• M.S. in Artificial Intelligence, University of Chinese Academy of Sciences — Hangzhou Institute for Advanced Study, 2024.09 – 2027.07 (expected)
  • Advisor: Prof. Jianqiang Huang (Computer Network Information Center, CAS)
  • GPA: 3.9/4.0 · Recommended admission (保研)
  • Honors: National Scholarship (2%), First-class Academic Scholarship (10%), Outstanding Teaching Assistant
• B.S. in Data Science and Big Data Technology, Nankai University, 2020.09 – 2024.07
  • GPA: 3.6/4.0 (3.84/4.0 in junior & senior years)
  • Honors: Outstanding Graduate (3%), Class Monitor leading the 12th Zhou Enlai Class (highest honor at Nankai)

Industry experience

• Algorithm Intern (Agent Post-training), Alibaba Group — Qwen Application Algorithm Team, Beijing, 2026.04 – present
  • Built automated agent evaluation pipeline (parsing, inference, metrics, Web GUI); delivered 8k+ train/eval samples at 97.8–100% accuracy, +10pp end-to-end eval usability
  • Designed rule-based + LLM-Judge layered rewards for agentic RL; +4pp parameter-call accuracy over SFT; supported Qwen3.5-122B launch (+1.96pp UV conversion in AB test)
• Algorithm Intern (Long-video Evaluation), Bilibili — Multimodal Foundation Model Team, Shanghai, 2025.12 – 2026.03
  • Proposed DSA metric with MLLM scoring; Spearman 0.76 vs. human judgment (paper: Long-CODE, ACM MM 2026 under review)
  • GRPO with SDE rollout for Wan2.1-1.3B; DSA improved 0.12 → 0.18
• Algorithm Intern (Multimodal Post-training), Shopee — Multimodal Foundation Model Team, Shanghai, 2025.06 – 2025.11
  • CompassVL-MoE-30B post-training via VeRL; OpenCompass 38.4 → 43.4, MathVista 77.9, OCRBench 83.2
  • Dual-reward RL for caption completeness & correctness; CCCaption-2B matches Qwen3-VL-32B (CVPR 2026, first author)

Research experience

• Multimodal Captioning & Evaluation, UCAS, 2025.03 – 2026.01
  • Semantic-equivalent image reconstruction for objective caption evaluation; unified multimodal model evaluation framework (2× NeurIPS 2026 under review)
• ECG Time-Series Benchmark, UCAS, 2024.10 – 2025.03
  • Built ECGPretrain (~15.9 GB) and 4-task benchmark (classification, detection, forecasting, generation); PSSM model with up to +67.2% over SOTA (ACM MM 2025)

Teaching

• Teaching Assistant, Natural Language Processing, UCAS, Fall 2024

Skills

• ML/AI: LLM SFT & RL (GRPO, DAPO), multimodal LLMs, agent evaluation, VeRL, vLLM, PyTorch, transformers
• Languages: Python, SQL, LaTeX
• Tools: Linux, Git, Claude Code

Publications

A Reconstruction-Based Framework for Caption Evaluation Beyond Reference Captions

Tang, Z., Qi, J., Tang, K., Zheng, Y., & Huang, J. (2026). "A Reconstruction-Based Framework for Caption Evaluation Beyond Reference Captions." NeurIPS 2026 (under review).

Do You See What You Draw? A Semantic Closed-Loop Framework for Holistic Evaluation of Unified Multimodal Models

Zhang, H., Qi, J., Tang, Z., & Huang, J. (2026). "Do You See What You Draw?" NeurIPS 2026 (under review).

Long-CODE: Isolating Pure Long-Context as an Orthogonal Dimension in Video Evaluation

Tang, Z., Qi, J., Zhao, B., & Huang, J. (2026). "Long-CODE: Isolating Pure Long-Context as an Orthogonal Dimension in Video Evaluation." ACM MM 2026 (under review).

CCCaption: Dual-Reward RL for Complete and Correct Image Captioning

Tang, Z., Wang, L., Qi, J., Jiang, W., Hou, P., Zeng, A., & Huang, J. (2026). "CCCaption: Dual-Reward RL for Complete and Correct Image Captioning." CVPR 2026.

A Comprehensive Benchmark for Electrocardiogram Time-Series

Tang, Z., Qi, J., Zheng, Y., & Huang, J. (2025). "A Comprehensive Benchmark for Electrocardiogram Time-Series." ACM MM 2025.

Radiation Reduction for Interventional Imaging via Video Frame Interpolation

Tang, Z., Xiong, Q., Wu, X., Xu, T., Shi, Y., Xu, X., Xu, J., & Wang, R. (2024). "Radiation Reduction for Interventional Imaging via Video Frame Interpolation." Insights into Imaging, 15(1).

Selected awards

• National Scholarship, UCAS (2024, top 2%)
• Challenge Cup National Second Prize (2023, top 1%)
• China Undergraduate Mathematical Contest in Modeling — National Second Prize (2022, top 2%)
• National College Student Innovation Training Program (top 10%), DSA medical imaging project