NUS · Computing / AI Systems
NUS PhD · Emerging AI systems researcher for executable agent infrastructure
Turning agent memory, execution semantics, and knowledge systems into a research program for reliable executable AI agents.
I am an NUS PhD student advised by Prof. Xiaokui Xiao, building the systems layer that makes uncertain AI execution stable and governable. My work sits where databases, agent runtime infrastructure, and machine learning systems meet: how an agent stores procedural knowledge, retrieves the right context, executes with transactional discipline, and improves from previous attempts.
I came to this problem through both research and production systems: GEM-Bench, my first-author SIGKDD 2026 CCF-A paper on generative engine marketing benchmarks; VDSAgents, a journal paper on PCS-guided multi-agent data-science automation; FOKE, my first-author work on personalized explainable education; and production experience optimizing infrastructure for Open-Sora 2.0, a former open-source SOTA text-to-video model, plus backend development for a commercial text-to-video platform serving 300,000 users. I was admitted to the NUS PhD program with a full scholarship through the NUSGRTII innovation and entrepreneurship track.
My current research agenda is deliberately narrow and compounding: AI-native databases for agents, latent state and procedural memory, verifiable multi-agent workflows, and decentralized capability invocation. On the systems side, I am building EasyRemote / EasyNet, an open-source network for agent execution and ability calls across devices. The long-term goal is a credible substrate for executable AI agent networks — a layer agents can use to remember, coordinate, and act.
silan-viking — the personal context system behind this site — has been making good progress lately. The Rust engine and CLI are coming together, with content sync and publishing now working end to end.
The first benchmark for evaluating how well an LLM can weave a sponsored ad into its answer — measuring whether the injection stays helpful, gets noticed, and survives user trust.
Authors: Silan Hu, Shiqi Zhang, Yimin Shi, Xiaokui Xiao
A five-agent pipeline (Define → Explore → Model → Evaluate, with a PCS guide) that automates an end-to-end data-science workflow without giving up reproducibility.
Authors: Yunxuan Jiang, Silan Hu, Xiaoning Wang, Yuanyuan Zhang, Xiangyu Chang
An open-source video diffusion model trained for $200k — proving you can hit commercial-grade quality on a startup budget by branching denoising trajectories and picking the best with a verifier.
Authors: Zangwei Zheng, Xiangyu Peng, Yuxuan Lou, Chenhui Shen, Tom Young, Xinying Guo, Binluo Wang, Hang Xu, Hongxin Liu, Mingyan Jiang, Wenjun Li, Yuhui Wang, Anbang Ye, Gang Ren, Qianran Ma, Wanying Liang, Xiang Lian, Xiwen Wu, Yuting Zhong, Zhuangyan Li, Chaoyu Gong, Guojun Lei, Leijun Cheng, Limin Zhang, Minghao Li, Ruijie Zhang, Silan Hu, Shijie Huang, Xiaokang Wang, Yuanheng Zhao, Yuqi Wang, Ziang Wei, Yang You
China National Conference on Big Data and Social Computing, pp. 399–411
A personalised education framework that fuses an LLM with a knowledge-forest graph of the learner, so explanations land at the right depth for the right student.
Authors: Silan Hu, Xiaoning Wang
Agent-driven query negotiation and a hierarchical, dynamic knowledge system for procedural memory.
GEM-Bench — the first benchmark for generative engine marketing, evaluating LLM ad injection, visibility, and user trust.
PCS-guided multi-agent system (Plan-Critique-Synthesize orchestration framework) for veridical data science automation.
TextRAG + GraphRAG hybrid knowledge system integrating foundation models, knowledge graphs, and prompt engineering.
Selected as the only AI-marketing team among 43; first to reach commercial closure. Part of the NRF S$50M GRIP program.
ScholarHero won first place in the intelligent teaching-assistant track, School of Data Science and Intelligent Media.
Faculty of Innovation Engineering; awarded for outstanding academic performance, conferred under the Dean's seal.
Faculty of Innovation Engineering; selected across multiple consecutive terms.
Project 'Starlink: Leading Smart Connectivity in the Industrial Internet Era' — reached the final 22 of 122 entries.
Project 'Pandemic Guardian'; supported by the Liaison Office and the Economic & Technological Development Bureau, serving 60+ universities and 4,000+ people across the GBA.
81 teams, ~300 students, 20+ countries — mostly graduate teams. Led an undergraduate team into the global top 5%.
MUST annual Cultural & Academic Excellence Award, first prize.
Capture-the-flag (CTF) competition.
