Xin Li李鑫
Portrait of Xin Li

About

I am a Ph.D. student at Nanyang Technological University (NTU), advised by Prof. Chau Yuen. I work on LLM agents: benchmarks that evaluate what they can do, training that improves them, and orchestration that gets real tasks done with tools, verifiers and other agents — and, lately, agents that improve themselves. So far most of the work has been in domains where an answer can be checked: formal verification, mathematics, and code. The current work is on self-improvement beyond checkable answers.

Before my Ph.D. I worked on robot perception — visual-inertial odometry at MEGVII, RGB-D + IMU indoor mapping at Microsoft Research Asia, and multimodal localization at Gausium Robotics, where I led a five-engineer team and shipped to a fleet of 1,000+ commercial cleaning robots — and then on LLM evaluation as a research assistant at NTU, building WirelessMathBench. During the Ph.D. I have interned at ByteDance (TikTok Shop), on the multimodal LLM review ranker, and at Microsoft Research Asia, on agent-generated slides.

Experience

2026 – now

ML Engineer Intern, ByteDance (TikTok Shop). Drove the multimodal LLM review ranker for TikTok Shop across global markets: ran its full post-training pipeline (SFT → RL → on-policy distillation) and took it through online A/B tests across regions with positive results; built an LLM agentic system that audits live review sections weekly and feeds the failures back into training.

2026

Research Intern, Microsoft Research Asia (remote). Agent-generated slides: rule-based checks so decks made by AI agents follow templates and rules. Paper: AesCode (arXiv 2026).

2024 – 2025

Research Assistant, NTU Singapore. With Prof. Chau Yuen. Built WirelessMathBench, the first benchmark for LLM mathematical modeling in wireless communications (ACL 2025 Findings), and TransPathNet for indoor radio-map prediction (ICASSP 2025).

2022 – 2024

SLAM Algorithm Engineer, Project Lead, Gausium Robotics, Singapore. Led five engineers on hierarchical multimodal localization — vision, LiDAR, Wi-Fi — for commercial cleaning robots: 95% localization accuracy across 100,000+ m², deployed on a fleet of 1,000+ active robots.

2020 – 2021

Research Intern, Microsoft Research Asia, Beijing. With Dr. Yang Liu and Dr. Yizhong Zhang. RGB-D + IMU fusion for large-scale indoor mapping; vectorized maps of 10,000+ m² at sub-meter accuracy.

2019 – 2020

Research Intern, MEGVII, Beijing. With Dr. Yijia He. Point-line visual-inertial odometry (IROS 2020); semi-dense 3D mesh reconstruction at 30+ FPS.

Education

2025 – now

Ph.D., Nanyang Technological University, Singapore (expected 2028). Advised by Prof. Chau Yuen.

2018 – 2021

M.E., Peking University. Advised by Prof. Jinlong Lin.

2014 – 2018

B.E., Northeastern University, China.

Grants & awards

2026

NeurIPS 2026 Top Reviewer

2026

Google Gemini Academic Program Award — US$10,000

2025

Modal Academics Compute Grant — US$2,000

2025

Cohere Labs Catalyst Grant — US$1,500

2025

OpenAI Researcher Access Program — US$1,000

2025

Rohde & Schwarz Award, IEEE 6G Summit Singapore

2025

PREMIA Best Student Paper Award, finalist

2025

NTU Research Scholarship, full Ph.D. funding

Mentoring

2026

Chengqi Liang — M.Sc. dissertation, Nanyang Technological University. Went on to a Ph.D. at The Chinese University of Hong Kong, Shenzhen.

2026

Yukun Jin — B.Eng. final-year project, Nanyang Technological University. An undergraduate at Wuhan University on NTU's 3.5+0.5+1 integrated programme; went on to an M.Sc. at NTU.

2024

Haoyu Xu — M.Comp. dissertation, National University of Singapore. Went on to a Ph.D. at Peking University.

Invited talks

Oct 2025

Teaching Large Language Models Mathematical Reasoning in Wireless Communications: From Benchmarking to Efficient Training — ISEE Youth Scholar Forum, School of Information and Electronic Engineering, Zhejiang University

Jun 2025

WirelessMathBench: A Mathematical Modeling Benchmark for LLMs in Wireless Communications — NICE Session 66, Evaluation Paper Sharing @ ICML & ACL 2025 (slides)

Service

Reviewer

NeurIPS (Top Reviewer, 2026), ICLR, ICML, AAAI, CVPR, ECCV, AISTATS, SIGGRAPH, IROS, ICRA · IEEE RA-L, ACM TOG, IEEE TNNLS

Organizer

AIR4D workshop @ IROS 2025