Portrait of Yishu Wang

Yishu Wang

王一澍

Product Manager II @ OKX

PhD candidate (on leave), HKUST Information Systems

Biography

I work at the intersection of LLM agents and forecasting — researching how AI systems gather information, conduct long-horizon reasoning, and eventually make decisions with real stakes. I am currently a Product Manager II at OKX, the world's second-largest cryptocurrency exchange by volume, where I am part of the Prediction Market team.

I am also a PhD candidate (on leave) in Information Systems at the Hong Kong University of Science and Technology. My research spans forecasting agents and AI for decision-making. I have published several papers at top conferences and their workshops, including ICML and IJCAI. When it comes to open source, I built Predict-Raven, an autonomous forecasting agent that trades independently on Polymarket.

Interests

  • LLM-based agents
  • Prediction markets
  • AI for decision-making
  • Agent evaluation

Education

  • Ph.D. in Information Systems, Hong Kong University of Science and Technology (on leave)
  • M.S. in Finance, Johns Hopkins University
  • B.Eng. in Computer Engineering, The University of Hong Kong

News

  • 2026Beyond Forecasting accepted at the ICML 2026 and IJCAI 2026 workshops.
  • Mar 2026Open-sourced Predict-Raven, an autonomous prediction-market trading agent — 360k+ views on social media.
  • Feb 2026Joined OKX as Product Manager II, leading the Agentic Wallet and Prediction Market products.
  • Fall 2025Teaching Assistant for Machine Learning at Johns Hopkins University.
  • 2025Reviewer for a NeurIPS 2025 workshop.

Publications

Workshop Papers

ICML 2026 Workshop · IJCAI 2026 Workshop

Yishu Wang, Yuxuan Wang, Hanyang Tang (2026). Beyond Forecasting: The Belief-to-Trade Layer in Prediction-Market Agents.

Manuscripts

Manuscript · 2026

Hanyang Tang, Yuhan Tang, Yishu Wang, Hongyi Lin (2026). Conditional Tabular Synthesis under Structured and Partial Urban Constraints.

Manuscript · 2026

Yishu Wang, Kakam Chong, Xiaofeng Wang, Xu Yan, DeXin Kong, Chen Ju, Ming Chen, Shuai Xiao, Shuguang Han, Jufeng Chen (2026). Evaluating Multi-Turn Bargain Skills in LLM-Based Seller Agent. Work during internship at Alibaba.

Projects

Predict-Raven

Autonomous forecasting agent framework that trades independently on Polymarket without human intervention.

  • Self-running agent that dynamically searches for information, evaluates event probabilities, and executes trades on prediction markets.
  • Both codebase and live trading records are fully open-sourced.
  • Achieved 360k+ impressions and 800+ likes/bookmarks on social media platforms.
Code on GitHub

Hack-Balatro

Training a superhuman AI model for the card game Balatro, with a fully simulated game environment and LLM-based evaluation harness.

  • Built a complete game simulation environment replicating all numerical values and game mechanics.
  • Currently using LLM and custom harness framework for evaluation while collecting SFT data.
  • Planning to explore train-free methods and RL approaches in subsequent phases.
Code on GitHub

Experience

Product Manager II OKX Feb 2026 – Present

Leading Agentic Wallet and Prediction Market products at the world's second-largest cryptocurrency exchange (110M+ users, ~$12T annual volume).

  • Agentic Wallet: Enabling agents to hold assets, interpret user intent, and execute transactions autonomously. 500k+ daily transactions, 40M+ daily API calls. Built automated testing system that improved agent transaction success rate from 91.3% to 99.7% and reduced false rejection rate by 35%.
  • Prediction Market: Self-developed prediction market with 100k+ DAU, 80k daily traders, 500k+ cumulative users. Managed full lifecycle operations, built on-chain data monitoring dashboards from 100M+ records.
Algorithm Engineer Intern Alibaba Group (Xianyu 闲鱼) Jul 2025 – Sep 2025

Built LLM evaluation benchmarks for e-commerce AI agents and deployed AI customer service workflows, reducing per-ticket processing time by 10%.

  • E-commerce Benchmark: Designed offline evaluation for AI-managed second-hand goods (1.5M+ managed items). Built multi-agent workflow for intent extraction with coverage improved from 0.72 to 0.89.
  • Workflow Optimization: Parallelized testing framework, improved KV-caching, reducing token consumption by 700M and test duration to 1/3.
  • AI Customer Service: Deployed LLM workflow reducing average handling time from 7.9min to 7.1min, saving ¥2.5M/year.
Product Manager (Full-time) Sortes.Fun Jan 2025 – Jul 2025

Led development of on-chain lottery service on Ethereum and Solana with verifiable voting for donation allocation. Secured $200k+ early-stage funding.

  • On-chain Lottery: Built lottery service with verifiable voting mechanism on Ethereum/Solana. $110k historical transaction volume.
  • Frontend Development: Built on-chain data indexing interface using Ether.js and Web3-onboard/React with Vite + TypeScript.
  • Product Design: Authored PRD for on-chain voting, designed backend data structures, coordinated dev team via Linear.
Product Intern ByteDance May 2022 – Aug 2022

Conducted VR/AR hardware research, user research for live-streaming products, and assisted with strategic investment due diligence.

  • VR/AR Research: Focused on digital asset creation and motion capture, analyzing headset device trends.
  • Live-streaming User Research: Conducted 50+ cold-call interviews and synthesized 20+ expert interviews.
  • Strategic Investment: Screened potential investment opportunities and assisted with Rec Room due diligence.

Service

Workshop Reviewer Conference on Neural Information Processing Systems (NeurIPS)
2025
Conference Organizer Advance Discussion in Blockchain Research, HK Web3 Festival
2023
Podcast Host Interview with Prof. Zhijing Jin (University of Toronto): Causal Inference and AI Safety

Teaching

Teaching Assistant – Machine Learning (BU.232.775)

Johns Hopkins University

Fall 2025

Demonstrated agent use cases, assisted professor with Python code deployment on Google Colab, provided Q&A support during lab sessions and seminars.

Instructor – Blockchain & Cryptocurrency

The University of Hong Kong

Spring 2023

Designed and taught a credited minor course in the Faculty of Engineering. Led curriculum design, content planning, and invited academic guests. 40+ enrolled students.