Interactive Memory & Recursive Self-Improvement (RSI) for World Models
Project Lead · World model evaluation, scalable data curation, training, self-improvement. Ongoing.
Ph.D. student · Renmin University of China
Building the data behind intelligent agents.
Scalable data synthesis, multimodal agents, and action-conditioned world models.
范梅浩Beijing, China
SCROLL TO DISCOVER PERSONAL WEBSITE / MEIHAO FAN
Seeking a Summer 2027 U.S. research internshipGet in touch
01 / A LITTLE CONTEXTData is where
Data is where
capability begins.
I am Meihao Fan (范梅浩), a Ph.D. student in Computer Application Technology at Renmin University of China, advised by Prof. Ju Fan and Prof. Xiaoyong Du. I completed my B.S. at Chongqing Jiaotong University in 2023.
My research focuses on scalable data synthesis for agent training. I develop methods for task generation, interaction-trajectory synthesis, and execution-based data verification, and use the resulting data for curriculum learning and multi-turn reinforcement learning. My work connects data construction to measurable improvements in agent capability and generalization. I am extending this research to embodied data curation, action-conditioned world models, and evaluation-driven data evolution.
I am seeking a Summer 2027 U.S. research internship in embodied AI data, world models, and multimodal agents.
02 / SELECTED WORK
From executable dataResearch in focus.
to more capable agents.
DeepPrep: Verifiable Task Synthesis for Agent Training
Project Lead; First Author · Executable supervision, reversible corruption, multi-turn RL. VLDB 2026.
Explore projectDeepAnalyze: Scalable Training-Trajectory Synthesis
Project Lead · Synthetic data, quality filtering, curriculum learning, agentic RL. ICML 2026.
Explore projectAutoPrep: Multi-Agent Data Preparation Framework
A multi-agent data preparation framework with planner and programmer agents.
Explore project03 / PUBLICATIONS
Google Scholar Ideas, in print.
Under review at ICLR 2027EvoCost: Harness Self-Evolution for Cost-Efficient LLM Agents
Meihao Fan, Shaolei Zhang, Ju Fan, Peng Li, Jie Song, Jianjun Chen
ICML 2026CoDA-Bench: Can Code Agents Handle Data-Intensive Tasks?
Yuxin Zhang, Ju Fan, Meihao Fan, Shaolei Zhang, Xiaoyong Du
ICML 2026DeepAnalyze: Agentic Large Language Models for Autonomous Data Science
Shaolei Zhang, Ju Fan, Meihao Fan, Guoliang Li, Xiaoyong Du
VLDB 2026TACO: A Benchmark for Open-Domain Text-to-SQL with Ambiguous and Cross-Database Queries
Chao Deng, Ju Fan, Yuyu Luo, Qinliang Xue, Meihao Fan, Yuxin Zhang, Min Zhang, Xiaofeng Jia, Jing Zhang, Xiaoyong Du
VLDB 2026DeepPrep: An LLM-Powered Agentic System for Autonomous Data Preparation
Meihao Fan, Ju Fan, Yuxin Zhang, Shaolei Zhang, Xiaoyong Du, Jie Song, Peng Li, Fuxin Jiang, Tieying Zhang, Jianjun Chen
SIGMOD 2026Reward-SQL: Boosting Text-to-SQL via Stepwise Execution-Aware Reasoning and Process-Supervised Rewards
Yuxin Zhang, Meihao Fan, Ju Fan, Mingyang Yi, Yuyu Luo, Jian Tan, Guoliang Li
VLDB 2025AutoPrep: Natural Language Question-Aware Data Preparation with a Multi-Agent Framework
Meihao Fan, Ju Fan, Nan Tang, Lei Cao, Guoliang Li, Xiaoyong Du
ICDE 2024Cost-Effective In-Context Learning for Entity Resolution: A Design Space Exploration
Meihao Fan, Xiaoyue Han, Ju Fan, Chengliang Chai, Nan Tang, Guoliang Li, Xiaoyong Du
Wireless Communications and Mobile ComputingFew‐Shot Multihop Question Answering over Knowledge Base
04 / THE JOURNEYLearning by
Learning by
building.
EvoPhys
Sept 2026 — PresentWorld model training, interactive-memory evaluation, multimodal data annotation, and evaluation-driven self-improvement.
ByteDance
March 2025 — June 2026DeepPrep: synthetic training tasks, executable data preparation, progressive SFT/RL, and agent harness self-evolution.
Renmin University of China
Sept 2023 — March 2025AutoPrep and BATCHER: task-aware data preparation, multi-agent orchestration, and cost-efficient LLM inference.