Hi, I'm Dequan.

Dequan Wang / 王德泉

Associate Professor at Shanghai Jiao Tong University

Full-time Mentor at Shanghai Innovation Institute

Making Intelligence Compound

I build agentic AI systems, AI-for-Science workflows, and evaluation loops that turn model capability into disciplined search.

DW

About

My work studies how intelligence compounds when models are placed inside persistent contexts, shared budgets, tool environments, and verification loops. I focus on agentic AI systems and AI for Science, with applications spanning autonomous agents, scientific benchmarks, protein and crop design, medical AI, and multimodal foundation models.

I build agentic harnesses for scientific swarms: systems that do not merely contain what we have discovered, but learn how to discover, test, compress, and carry forward their own experience.

Positions

SJTU
Shanghai Jiao Tong University
Associate Professor
2026 - Present
SII
Shanghai Innovation Institute
Full-time Mentor
2025 - Present
SAIL
Shanghai Artificial Intelligence Laboratory
Research Scientist
2023 - 2024
SJTU
Shanghai Jiao Tong University
Assistant Professor
2023 - 2025

Education

UCB
University of California, Berkeley
Ph.D. in Computer Science
2016 - 2022
FDU
Fudan University
B.S. in Computer Science
2012 - 2016

Research Stack

Agentic AI
Agent Harnesses
Multi-Agent Systems
Evaluation
Verification
Long-Context Agents
AI for Science
Protein Design
Crop Engineering
Medical AI
Foundation Models
Multimodal AI
Scientific Benchmarks
Featured Work

Selected systems, benchmarks, datasets, and scientific AI projects.

A curated subset of work that represents the current research direction.

AgencyBench
arXiv

AgencyBench

Benchmarking autonomous agents in long, real-world contexts.

Agentic AI
Benchmark
Long Context
Evaluation
Credit-Budgeted ICPC-Style Coding
arXiv

Credit-Budgeted ICPC-Style Coding

Evaluating coding agents under explicit reasoning and action budgets.

Agentic AI
Coding
Evaluation
Budgeted Reasoning
Aligned Agents, Biased Swarm
arXiv

Aligned Agents, Biased Swarm

Studying bias amplification and failure modes in multi-agent systems.

Multi-Agent Systems
Bias
Alignment
Evaluation
MAC
arXiv

MAC

A live benchmark for multimodal large language models in scientific understanding.

Multimodal AI
Scientific Understanding
Benchmark
AI4Protein
Paper

AI4Protein

A survey and perspective on AI-driven protein design.

AI for Science
Protein Design
Scientific AI
Compact Enhancers for Crop Engineering
Paper

Compact Enhancers for Crop Engineering

AI-guided discovery and design of compact enhancers for crop engineering.

AI for Science
Crop Engineering
Genomics
MedFMC
arXiv

MedFMC

A real-world dataset and benchmark for foundation model adaptation in medical image classification.

Medical AI
Foundation Models
Dataset
Benchmark
Back to the Source
arXiv

Back to the Source

Diffusion-driven adaptation to test-time corruption.

Robustness
Test-Time Adaptation
Diffusion
Recent Signals

Recent public updates and milestones.

SII

智能体的“马拉松教练”藏在GitHub里

上海创智学院

Public article from Shanghai Innovation Institute on agentic systems and long-horizon evaluation.

Article
SII

评测永无止境，其边界决定了智能的上限

上海创智学院

Public article from Shanghai Innovation Institute on evaluation boundaries and intelligence.

Article
Contact

Get in Touch

For research, teaching, and collaboration inquiries, email is the most direct route. dequanwang@sjtu.edu.cn

GitHub
LinkedIn
X