Hi, I’m Allan.

I’m a Research Scientist at ByteDance Seed in Singapore, working on LLM agents and reinforcement learning. I’m a core contributor to Seed-Prover, a family of Lean 4 theorem provers trained with large-scale agentic RL — Seed-Prover reached an IMO-certified, silver-medal-level score at IMO 2025, and Seed-Prover 1.5 proves 88% of PutnamBench.

I’m especially interested in agentic RL: long-horizon credit assignment, reward design, and learning from experience.

Before Seed, I was a Research Scientist at Salesforce AI Research, working on efficient long-context inference (ThinK, an ICLR 2025 Spotlight), and an NLP Scientist at ByteDance AI Lab, working on mathematical reasoning with language models — including ReFT, which brings reinforcement learning to chain-of-thought fine-tuning.

I received my Ph.D. from the Singapore University of Technology and Design, advised by Prof. Wei Lu in the StatNLP group, where I built structured models for named entity recognition and semantic parsing; the thesis received SUTD’s Best Thesis Award. Along the way I interned at the Allen Institute for AI and at Alibaba.