cv
Contact Information
| Name | Bo Zhang |
| Professional Title | Undergraduate Researcher |
| bozhang23@mails.tsinghua.edu.cn |
Professional Summary
Undergraduate researcher at Tsinghua University working on multimodal agentic systems — building agents that act reliably across pixels, simulations, and classrooms.
Education
-
2023 - 2027 Beijing, China
B.Eng.
Tsinghua University
Department of Automation
- Research focus on multimodal agentic systems, computer-use agents, and AI for science.
Publications
-
2026 BAMI: Training-Free Bias Mitigation in GUI Grounding
CVPR 2026
Training-free inference-time method that combines coarse-to-fine focus with candidate selection to mitigate precision and ambiguity biases in GUI grounding. Lifts TianXi-Action-7B accuracy on ScreenSpot-Pro from 51.9% to 57.8%. Co-first author (order decided by coin flip).
-
2026 Grounding LLMs in Scientific Discovery via Embodied Actions
ICML 2026
EmbodiedAct grounds LLMs in scientific software via a tight perception–execution loop. Instantiated within MATLAB; achieves SOTA on engineering design and scientific modeling with strong long-horizon stability. Co-first author.
Awards
-
2025 Grand Prize (特等奖) — 43rd Tsinghua Challenge Cup, AI Teaching-Agent Track
Tsinghua University
Project AEI4U — an AI teaching assistant for Fundamentals of Analog Electronics built on the Dify-THU platform. Project Lead.
Projects
-
AICosmos
Educational multi-agent platform incubated by Tsinghua’s Department of Computer Science. Tutor / Role-Play / Q&A agents composed into a course-grade assistant. Live at aicosmos.ai.
- Multi-agent orchestration
- Educational and research workflows
-
Syll
Self-hosted, MIT-licensed AI companion with a web UI, multi-channel chat, proactive rituals, markdown skills, and a desktop ghost. Core contributor at THU-SAGE/syll.
- Recorded-workflow studio
- Cross-channel (Telegram, Feishu, Discord, WhatsApp, Web, CLI)
- PyQt6 desktop ghost
-
AEI4U
AI teaching assistant for analog electronics on Dify-THU. Project Lead. Won Grand Prize at Tsinghua’s 43rd Challenge Cup AI Teaching-Agent track.
- Circuit topology recognition
- SPICE-in-the-loop simulation
- Adaptive learner profiling