English | 中文
Open to full-time roles · from early 2027

Yanjie Zhang

I study how AI agents resolve conflicting information: when a user retracts an earlier instruction, when memory disagrees with what the agent now observes, when what is heard disagrees with the transcript, or when a chart's visuals contradict its data.

I am an MSc student in Big Data Technology at HKUST and a Machine Learning Research Intern at NVIDIA NVAITC, collaborating with the Healthcare AI Team on medical AI and Nemotron post-training. Previously, I worked at Tencent Games and Binance, where I was a core contributor to Binance AI and Binance AI Pro.

I am open to AI research and algorithm roles, AI-focused investment research, and PhD positions. Feel free to reach out!

Yanjie Zhang

Publications

26 citations · h-index 3 (Google Scholar). * denotes equal contribution.

MIRAGE
Navigating the Mirage: A Dual-Path Agentic Framework for Robust Misleading Chart Question Answering
Yanjie Zhang, Yafei Li, Rui Sheng, Zixin Chen, Yanna Lin, Huamin Qu, Lei Chen, Yushi Sun
ACM Multimedia 2026 (CCF-A) · A dual-path agent framework; SFT + GRPO lifts Qwen3-VL-8B to 74.43% accuracy.
Visuals ↔ dataarXiv
VizDefender
VizDefender: Unmasking Visualization Tampering through Proactive Localization and Intent Inference
Sicheng Song, Yanjie Zhang, Zixin Chen, Huamin Qu, Changbo Wang, Chenhui Li
IEEE TVCG / PacificVis 2025 (CCF-A) · Honourable Mention Award
Tampered ↔ authenticarXiv
IntentFlux
When Users Change Their Minds: Measuring and Repairing Intent Drift in LLM Agents
Yanjie Zhang, Bowen Cao, Zixin Chen, Yushi Sun
Under review, ICLR 2027 · IntentFlux benchmark and the StateForge runtime harness.
Instruction ↔ instructionarXiv
VGBench
Do Audio LLMs Listen Before They Act? Diagnosing Acoustic-Context Gating in Voice Agents
Yanjie Zhang, Nanchen Hu, Yushi Sun
Under review, ICLR 2027 · VGBench and a post-training mitigation.
Audio ↔ transcriptarXiv
When Memory Lies
When Memory Lies: An Empirical Study of Spatial Memory Staleness in VLM Agents
Yushi Sun*, Yanjie Zhang*
Under review, AAAI 2027 (Phase 2) · *Co-first authors
Memory ↔ observationarXiv
Personalization Mirage
The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads
Yushi Sun*, Yanjie Zhang*, Rui Sheng
Under review, AAAI 2027 (Phase 2) · *Co-first authors
Belief ↔ evidencearXiv
ADAPT
ADAPT: Adaptive Lesson Plan Transformer for Cross-Regional and Differentiated Instruction
Yanjie Zhang, J Zhu, M Wu, H Qu, S Song
arXiv preprint
InconLens
InconLens: Interactive Visual Diagnosis of Behavioral Inconsistencies in LLM-based Agentic Systems
S Yan, X Wen, S Ruan, Yanjie Zhang, J Mi, Y Sun, H Qu, R Sheng
arXiv preprint
Inconsistent behaviorsarXiv
MDT Room
Within the MDT Room: Situated in Multidisciplinary Team-Grounded Agent Debate for Clinical Diagnosis
P Kuai, Y Yang, S Ruan, J Xu, Yanjie Zhang, L Zhang, M Zhu, R Sheng
arXiv preprint
MAS in Medicine
A Survey of LLM-based Multi-agent Systems in Medicine
Y Lin, S Xu, W Zhang, Y Sun, Z Chen, Yanjie Zhang, R Sheng
Survey

Industry Experience

NVIDIA · NVAITCMachine Learning Research Intern

Designing medical-AI benchmarks that expose hallucinations, and exploring SFT / RL post-training to strengthen Nemotron on clinical tasks.

Tencent Games · LightSpeed StudioResearch Scientist Intern

Studied how agents handle conflicting instructions, audio cues and memories (intent drift, acoustic-context gating, stale memory), building benchmarks plus training and runtime methods to resolve them.

BinanceAI Engineer

Built the multi-agent infrastructure, long-term memory, modular skill architecture and tracing / evaluation stack behind Binance AI.

Tencent Cloud · Network ProductsBackend Development Intern

Developed a Kubernetes-based NFV management platform with cross-AZ Pod failover, and sped up large-scale container image delivery.

HelloBike · Two-Wheeler Product CenterBackend Development Intern

Built high-concurrency microservices for overseas payments and orders, plus an automated risk-inspection platform.

Education

The Hong Kong University of Science and Technology QS 332025 – 2027

MSc in Big Data Technology · GPA 4.0/4.3

East China Normal University 9852020 – 2025

BSc in Computer Science and Technology · GPA 3.7/4.0

HKUST VisLabVisiting Research Intern & RA

Advised by Prof. Huamin Qu. Worked on robust chart understanding, visualization-tampering detection, and AI feedback for a VR English-learning platform.

Skills

LLMs & AgentsMLLMs, agent memory, multi-agent orchestration, SFT, PPO, GRPO, OPD, LLM evaluation

ProgrammingPython, TypeScript, Go, Java, C/C++, SQL, Solidity

SystemsKubernetes, Docker, Node.js, Spring Boot, Redis, MySQL, Kafka, Qdrant, Neo4j