Alongside empirical research, I build tools and investigate agent systems for working with scientific literature. This gives me a practical setting for thinking about how knowledge is accessed, retained, and used—and how AI assistance can be evaluated.
Frowang
·
An agent-native platform for literature research
Independent project · Native Agent Track Champion, EvoMap Hackathon 2026
I designed and built Frowang around a reading workflow: screen papers with structured summaries, inspect the original evidence, and use an assistant for closer reading. Paper APIs, a CLI, and installable skills connect local agents to the same library for retrieval, organization, and cross-paper work.
453 registered users · approximately 100 monthly active users · October 24, 2026
EvoMap
· Agent Memory & Agent Swarm
Algorithm Intern · October 2026–present
I work on agent memory systems and investigate swarms for literature discovery and synthesis. The swarm work asks how parallel retrieval and reading can reconstruct a research direction’s development when its literature exceeds one agent’s context, balancing coverage and relevance.
Yupu Hao,
Zhuoran Jin
,
Zichen Wang
, Jiachun Li, Kejian Zhu, Lingshuai Wang, Kang Liu, and Jun Zhao
Accepted · LASS Workshop at ACM CIKM 2026
·
Paper
SocialSim-Bench studies strategic interaction and simulation-based decision support. I contributed to negotiation scenarios and benchmark tasks, translating private goals and economic constraints into settings for evaluating agents’ behavior.
Abstract
Most prior LLM-based social-simulation studies ask whether agents appear realistic, socially intelligent, or behaviorally consistent. However, can LLM agents genuinely reason about evolving social dynamics and develop adaptive strategies, rather than merely role-play? Moreover, social simulations are often intended to provide guidance for real-world decision-making. We study the next step: how simulated possibilities can inform disciplined decision support. We present SocialSim-Bench, a benchmark and decision-assistant report-generation framework for multi-agent decision support, comprising two scenarios: social-media interaction and long-term negotiation. Each scenario combines contextual backgrounds, role-specific profiles, private goals, and action constraints. We evaluate agents using rule-based metrics and an LLM-as-a-judge system, while also analyzing the strategies they employ, finding that even the excellent LLMs only perform simple strategies without applying complex methods. We further investigate multiple approaches for transforming simulation results into decision-assistant reports, including direct LLM generation, harness-based generation with agents such as Codex, and our proposed recursive synthesis approach, which transforms simulation traces into decision-assistant reports by summarizing individual trajectories, aggregating recurring conditions and plausible outcomes, and synthesizing actionable recommendations. We evaluate report quality across multiple dimensions. Our results show that simulation-grounded reporting improves decision-support quality in both social-media and negotiation settings compared with no simulation.
Abstract
Most prior LLM-based social-simulation studies ask whether agents appear realistic, socially intelligent, or behaviorally consistent. However, can LLM agents genuinely reason about evolving social dynamics and develop adaptive strategies, rather than merely role-play? Moreover, social simulations are often intended to provide guidance for real-world decision-making. We study the next step: how simulated possibilities can inform disciplined decision support. We present SocialSim-Bench, a benchmark and decision-assistant report-generation framework for multi-agent decision support, comprising two scenarios: social-media interaction and long-term negotiation. Each scenario combines contextual backgrounds, role-specific profiles, private goals, and action constraints. We evaluate agents using rule-based metrics and an LLM-as-a-judge system, while also analyzing the strategies they employ, finding that even the excellent LLMs only perform simple strategies without applying complex methods. We further investigate multiple approaches for transforming simulation results into decision-assistant reports, including direct LLM generation, harness-based generation with agents such as Codex, and our proposed recursive synthesis approach, which transforms simulation traces into decision-assistant reports by summarizing individual trajectories, aggregating recurring conditions and plausible outcomes, and synthesizing actionable recommendations. We evaluate report quality across multiple dimensions. Our results show that simulation-grounded reporting improves decision-support quality in both social-media and negotiation settings compared with no simulation.