2026 — Present
Current
Nanyang Technological University
Ph.D. in Computer Science · College of Computing and Data Science
Advisor: Prof. Yang Liu · Singapore
LLM Post-training · Coding Agents · Agent Memory
Training agents to learn, remember, and improve through interaction.
I am Ziyang Zhou (Deric), a Ph.D. student in Computer Science at Nanyang Technological University, advised by Prof. Yang Liu. I received my B.Sc. from Xi'an Jiaotong-Liverpool University in 2026.
My current interests are LLM post-training, coding agents, and agent memory. I study how agents can learn from trajectories, preserve useful experience, and continually improve on long-horizon tasks.
01 / Education
Academic path
From a broad computer science foundation to focused doctoral research on learning agents.
2022 — 2026
Completed
Xi'an Jiaotong-Liverpool University
B.Sc. in Information and Computer Science
GPA 3.7 / 4.0 · Suzhou, China
ACL 2026 · Main Conference
RAM-SD accepted to ACL 2026 Main.
View publication ↗ICASSP 2026
Following the TRACE accepted to ICASSP 2026.
View publication ↗02 / Research
Selected publications
Work on agentic reasoning, affective language understanding, and multi-agent collaboration. Full list on Scholar ↗
03 / Experience
Selected experience
Work spanning LLM post-training, coding agents, agent memory, and production agent systems.
EigenAI
LLM Researcher
Developed data-centric post-training workflows for long-horizon web agents, including curriculum task generation and validator-based filtering.
Gerstein Lab
Deep Research & Self-Evolving Agents
Worked on cross-framework agent memory, coding-agent evolution, and reinforcement learning for self-improving agents.
Lab ↗
AxiomsTen
Algorithm Engineer Intern
Built agent services, Graph-RAG abstractions, and multi-turn reasoning workflows for personalized user interaction.
OPPO Personal AI
LLM Agent Collaborator
Designed retrieval-augmented agent memory and knowledge bases, improving GAIA performance by 18.7%.
Human-Centered Language Computing Group · XJTLU
Multi-Agent Language Research
Developed CAF-I, SEVADE, and RAM-SD for multi-agent reasoning, self-evaluation, and retrieval-augmented language understanding.
Audio Intelligence Lab · XJTLU
Audio Intelligence Research
Built device-aware acoustic scene classification and HuBERT-based accent recognition systems.
04 / Research direction
Useful agents should not only act. They should learn from what happened.
I am interested in post-training systems that turn trajectories into reusable memory, improve coding behavior, and help agents adapt across long-horizon tasks.
Let's talk about research ↗