Zhen Peng

Senior AI/ML expert and marketplace algorithm leader working on large-scale decision systems.

I lead supply-demand adjustment strategy for the ride-hailing marketplace at DiDi Global, where my team builds algorithms for marketplace optimization, dynamic pricing, intelligent subsidies, growth marketing, and AI-assisted operations. My work combines causal inference, reinforcement learning, forecasting, operations research, and LLM agents to improve real-time and long-horizon business decisions.

Before DiDi, I worked on distributed computing infrastructure at Baidu, where I was a founding engineer of the Baidu Spark computing engine and helped scale Hadoop/Spark systems across thousand-node clusters. Across industry and research collaborations, I focus on turning rigorous modeling into production systems that operate at marketplace scale.

Research interests Selected work Full CV Google Scholar

News

  • 2025 Paper accepted at CIKM on augmenting limited and biased RCTs with observational data fusion.
  • 2025 AAAI oral paper on decision-focused fine-tuning for predict-then-optimize systems.
  • 2024-2025 DiDi ride-hailing and CTO-line projects recognized with multiple outstanding project awards.
  • 2024 Recognized as an outstanding reviewer by DiDi's Algorithm Committee.

Research

My applied research centers on algorithmic decision-making in two-sided marketplaces: estimating price elasticity and supply-demand dynamics, designing multi-objective pricing and subsidy mechanisms, optimizing long-term user value, and integrating LLM agents into human-in-the-loop decision workflows.

I collaborate with university labs including Tsinghua University and Peking University on marketplace optimization, causal inference, reinforcement learning, and decision-focused learning, with publications at AAAI and CIKM and several productionized research systems.

Interests

  • Marketplace optimization
  • Dynamic pricing and revenue management
  • Causal inference and elasticity modeling
  • Reinforcement learning for decision systems
  • Forecasting and operations research
  • LLM agents and AI copilots
  • Distributed ML infrastructure

Selected Work

Augmenting Limited and Biased RCTs through Pseudo-Sample Matching-Based Observational Data Fusion Method

Han, H, ..., Peng Zhen, et al. CIKM, 2025.

DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with Limited Data

Yang, J., ..., Peng Zhen, et al. AAAI, 2025. Oral.

Contrastive Balancing Representation Learning for Heterogeneous Dose-Response Curves Estimation

Zhu, W, ..., Peng Zhen, et al. AAAI, 2024.

Code

My engineering work spans real-time pricing and subsidy decision engines, offline ML pipelines, experiment and evaluation platforms, feature systems, strategy workbenches, LLM application infrastructure, RAG/NL2SQL systems, and multi-tool Agent products.

Representative internal systems process hundreds of millions of online decisions per day and support intelligent planning over large-scale annual marketplace budgets.