Final-year Ph.D. candidate @ HKUST · I build AI systems that ship to production.
Agentic & ML systems for high-stakes, real-world decision-making — urban, transportation, and scientific domains.
Homepage · Scholar · X / Twitter · lyutf@outlook.com
I don't just publish models — I ship them. My work runs live, at scale, with verified gains.
- 🚕 ProfiLLM — live on DiDi (Brazil). An agentic LLM framework with utility-aligned user profiling for industrial ride-hailing dispatch. Deployed across three cities in Brazil, serving millions of daily orders, with verified live A/B-test gains.
🌐 Project · 💻 Code · (VLDB, under review) - ⚙️ DispatchEvolve — live on DiDi (4 cities). The first framework to bring LLM-driven program evolution to industrial dispatch engines: it autonomously discovers optimization opportunities from decision traces, evolves the implicated policies, and assembles them into complete engines under six coupled business objectives. Deployed across four cities, the only method to improve all six objectives in every city offline, with live A/B gains up to +1.32% GMV and +1.07% completion rate.
🌐 Project · 💻 Code · (KDD 2027 ADS, under review) - ☢️ NRFormer — deployed nationwide. A spatio-temporal transformer for nationwide nuclear-radiation forecasting, online for 1–24 day forecasts.
📄 arXiv · 💻 Code · (KDD 2025, CCF A)
I'm a final-year Ph.D. candidate in the AI Thrust at The Hong Kong University of Science and Technology, Guangzhou, advised by Prof. Hao Liu and Prof. Hai Yang (Chair Professor). I work closely with Prof. Hui Xiong (Chair Professor) on LLMs and foundation models for urban general intelligence.
What I build: trustworthy, efficient, deployable AI for decision-making — from autonomous LLM agents that route ride-hailing fleets, to physics-informed transformers that forecast climate and radiation nationwide.
🎯 I'm on the job market — seeking academic and industry positions. Let's talk: lyutf@outlook.com
- 🏆 5 published first-author CCF-A papers (KDD, IJCAI, SIGIR)
- 📚 2 first-author IEEE Transactions papers (TITS, TCBB)
- 📝 5 first/co-first author papers under review (VLDB, TKDE, KDD ×2, NMI)
- 📈 600+ citations (Google Scholar)
- 🌍 3 real-world deployments — ProfiLLM & DispatchEvolve on DiDi Brazil; NRFormer online nationwide
My work turns frontier AI into deployed systems that make consequential decisions in the physical world.
📂 Expand for per-paper details
🤖 LLM Agents for Decision-Making
- ProfiLLM — agentic LLM with utility-aligned user profiling for industrial ride-hailing dispatch; deployed on DiDi. Project · Code
- DispatchEvolve — autonomous LLM-driven policy evolution for production dispatch engines under six coupled business objectives; deployed on DiDi across four cities. Project · Code
- LLM-ODDR — the first LLM framework for joint order dispatching and driver repositioning. arXiv · Code
🔬 AI for Science
- TelePiT — Physics-Informed Teleconnection-Aware Transformer for global subseasonal-to-seasonal forecasting. arXiv · Code
- NRFormer — spatio-temporal transformer for nationwide nuclear-radiation forecasting, deployed online. arXiv · Code
- NRFormer+ — atmospheric diffusion-guided spatio-temporal transformer extending NRFormer for nuclear-radiation forecasting. arXiv · Code · (TKDE, under review)
- MDNN — multimodal deep neural net for drug–drug interaction events. Paper
- MGNN — multimodal GNN for cancer patient survival. Paper
⚙️ Automated Machine Learning (AutoML)
- 2026.07 ⚙️ Released DispatchEvolve — autonomous LLM-driven policy evolution for industrial dispatch engines, deployed on DiDi across four cities.
- 2026.06 🏆 Best Research Award (First place), AI at HKUST Guangzhou.
- 2026.05 🚕 Released ProfiLLM — agentic LLM for ride-hailing dispatch, deployed on DiDi Brazil.
- 2026.05 🎉 LLM-ODDR accepted to IEEE TITS.
- 2025.11 🎉 TelePiT accepted to KDD 2026 (global S2S forecasting).
- 2025.09 🥳 Paper accepted to NeurIPS 2025 (foundation models for scientific discovery).
- 2025.07 🏅 KDD Student Travel Award.
- 2025.05 🎉 NRFormer accepted to KDD 2025 (nationwide nuclear radiation forecasting).
- 2024.11 🎉 AutoSTF accepted to KDD 2025 (automated spatio-temporal forecasting).
Reviewer — SIGKDD (2024/2025/2026), WebConf (2025/2026), NeurIPS (2023/2024), TKDE, TKDD, TIST, TITS.
Selected awards — Best Research Award (2026, AI@HKUST Guangzhou) · ACM SIGKDD Student Travel Award (2025)· ACM SIGIR Student Travel Award (2020) · Outstanding Paper Award (2024) · Outstanding Graduate of Hunan Province, Top 1% (2022) · National Scholarship, Top 1% (2016/2018/2020) · Liaoning Provincial Government Scholarship, Top 1% (2017)· Presidential Scholarship, Top 0.1% (2018) · President Innovation Scholarship, Top 0.1% (2021).
📬 Building deployable AI for real-world decision-making — and looking for my next role.
Homepage ·
Scholar ·
X / Twitter ·
lyutf@outlook.com