Looking for Research Internships in LLM Memory & Agents
PhD student at The Hong Kong Polytechnic University, jointly trained with GDIIST (Guangdong Institute of Intelligence Science and Technology), working on brain-inspired memory, multi-agent systems, and recursive self-improvement — looking for an industry or lab internship where that work meets real systems. Remote roles worldwide, including US-based teams, are very welcome.
Overview
I am a PhD student at The Hong Kong Polytechnic University (PolyU), enrolled in a joint doctoral programme co-supervised with GDIIST (Guangdong Institute of Intelligence Science and Technology), working on large language models — brain-inspired memory architectures, multi-agent systems, and, increasingly, recursive self-improvement: how an agent can convert its own experience into lasting capability. I am currently looking for a research internship. With publications at EMNLP and KDD, and prior research at GDIIST, HKMU, Tsinghua NLP Lab, and Zhipu AI, I work across the full LLM stack — from multi-node distributed training to agent-level system design — and I am comfortable owning a research problem end to end, from literature to running code to a paper.
Academic Background
B.Sc. Information Management & Information Systems, Beijing University of Chinese Medicine (2018–2022)
Current Position
PhD Student, The Hong Kong Polytechnic University (PolyU), Hong Kong SAR — 2026–Present. Joint doctoral training with GDIIST (Guangdong Institute of Intelligence Science and Technology).
Research Experience
PolyU (2026–Present), GDIIST (2025–2026), HKMU (2023–2025), THU NLP (2023), Zhipu AI (2022–2023)
Research Focus
Brain-inspired memory, multi-agent systems, recursive self-improvement, RAG, reasoning, domain-specific LLMs
Community Involvement
Hugging Face Chinese Community Lead, AITIME Volunteer, Journal Reviewer (NLP Journal, IEEE IoT)
Technical Skills
Python, PyTorch, React, Docker, Smolagents, distributed training (H100 GPUs)
What I'm Looking For
Type of Internship
Research intern / student researcher at a large tech company's AI lab, an AI research institute, or a startup research team — ideally where the work can lead to a paper, an open-source release, or a shipped system. Research-engineer roles that sit between the two are equally welcome, and fully remote positions work well for me.
Topics I'd Love to Work On
Long-horizon memory for LLM agents, agent frameworks and evaluation, recursive self-improvement / self-evolving systems, RAG and reasoning, post-training and data pipelines for agentic models.
What I Bring
Publications at ACL, EMNLP (×2) and KDD, with two AAAI papers under review and an ICLR submission in preparation; benchmark and dataset construction from scratch, multi-node training on 20+ H100 GPUs, agent framework engineering, and full-stack delivery (Python, PyTorch, Docker, React). Also a reviewer for NLP Journal and IEEE IoT Journal.
Timing & Location
Based in Hong Kong. On-site: Hong Kong, Mainland China, Singapore. Remote: worldwide, including US and European teams — I already work across time zones and can overlap with US hours. Duration and start date are flexible; happy to fit both the team's cycle and my doctoral schedule.
Research Timeline
PhD Student — The Hong Kong Polytechnic University × GDIIST (joint doctoral programme)
Doctoral research on long-horizon memory for LLM agents, multi-agent coordination, and recursive self-improvement — agents that reflect on their own trajectories and revise their own strategies under verification
Research Assistant — GDIIST, Zhuhai
Brain-inspired memory architectures, Pain LLM with 20+ H100 GPUs, pain mouse behavioral benchmarks, China Brain Project, FDCT grant proposals
Research Assistant — HKMU, Hong Kong
CondAmbigQA benchmark (EMNLP 2025), condition reasoning improving LLM accuracy by 18.9%, GRF/NSFC grant proposals, journal reviewer
Research Collaborator — Tsinghua NLP Lab
XAgent framework for multi-agent collaboration, ReproAgent RAG module
Algorithm Engineer (Intern) — Zhipu AI (THU KEG)
CodeGeeX code generation model (KDD 2023), GLM-based poetry generation, prompt engineering
Founder Member — Shenzhen InnoX Academy
Emotional companion robots, SLAM-based autonomous navigation, smart wheelchairs, embodied AI
PhD Research Agenda
The through-line of my doctoral work is simple to state and hard to do: memory is the substrate of self-improvement. An agent that cannot durably remember what worked cannot reliably get better. The directions below run from what exists today toward what is still exploratory; the later ones are intentionally described at the level of direction rather than method.
Recursive Self-Improvement & Self-Evolving Agents
ExploratoryStudying how an agent can close the loop between experience and capability: accumulating durable experience, evaluating its own trajectories, revising its own strategies, prompts, and tooling, and compounding those revisions over time. The open questions I care about are what makes such a loop converge rather than drift, how improvement is measured when the evaluator is also changing, and what verification envelope keeps self-modification bounded and auditable. Directional, early-stage work — specifics deliberately not detailed here.
Brain-Inspired Memory Architectures for LLMs
Building on the BMAM framework to develop neuroscience-principled memory systems that enable agents with human-like consolidation, forgetting, and retrieval capabilities.
Multi-Agent Evaluation Frameworks
Developing robust methodologies to evaluate collaboration, specialization, and information sharing between multiple LLM-based agents in complex task environments.
Context-Aware Reasoning for Ambiguous Tasks
Building on CondAmbigQA research (EMNLP 2025) to enable LLMs to handle ambiguity through contextual understanding and multi-step reasoning processes.
Domain-Specific LLMs for Healthcare
Extending Pain LLM research to broader healthcare applications, leveraging experience with distributed training on large GPU clusters and behavioral benchmarking.
RAG Architecture for Specialized Domains
Exploring retrieval-augmented generation techniques optimized for domain-specific knowledge integration, focusing on medical and scientific applications.
Publications
Five accepted papers across ACL, EMNLP (×2), KDD, and a journal venue, with two AAAI submissions currently under review and an ICLR submission in preparation.
- 2× EMNLP (Main)
- 1× ACL (Findings)
- 1× KDD
- 1× Journal (JTCMS)
- 2× AAAI submissions
- On LLM memory and agent self-evolution — titles withheld until decisions are out
- 1× ICLR submission
- Currently in preparation
AURA: Intent-Directed Probing for Implicit-Need Surfacing in Situated LLM Agents
Y. Li, J. Liu, J. Cai, and M. Xu. arXiv:2606.05557
Adds an inference layer between scene perception and tool execution: an IntentFrame assesses the unstated needs behind a situated query, and a gap score decides how much probing each query deserves — broader implicit-need coverage than ReAct-style probing, with 82% fewer probes on factual lookups.
BMAM: Brain-inspired Multi-Agent Memory Framework
Y. Li, J. Liu, Y. Wang, Y. Wu, and M. Xu. Findings of ACL 2026. arXiv:2601.20465
CondAmbigQA: A Benchmark and Dataset for Conditional Ambiguous Question Answering
Z. Li*, Y. Li*, H. Xie, and S. J. Qin. pp. 2269–2288. *Equal contribution
CodeGeeX: A Pre-Trained Model for Code Generation with Multilingual Benchmarking on HumanEval-X
Q. Zheng, X. Xia, X. Zou, ..., Y. Li, ..., and J. Tang
Machine Learning Models for Stroke Detection by Observing Eye-Movement Features Under Five-Color Visual Stimuli in Traditional Chinese Medicine
Q. Lu, J. Deng, Y. Yu, Y. Li, et al. Journal of Traditional Chinese Medical Sciences, 10(3), pp. 321–330
Academic Service
Journal Reviewer
Natural Language Processing Journal (2024–2025)
Journal Reviewer
IEEE Internet of Things Journal (2025)
Community Lead
Hugging Face Chinese Community (2022–Present)
Community Volunteer
AITIME (2021–Present)
Contact for Internships & Collaboration
If your team is hiring research interns — on-site or fully remote, anywhere from Hong Kong to the US — or if you simply work on LLM memory, multi-agent systems, self-improving agents, or domain-specific models, I'd be glad to hear from you. A CV and a longer research statement are available on request.
Email: innovation64feng@gmail.com
Phone: +86 13269183099 / +852 54614337
Current Institution: The Hong Kong Polytechnic University (PolyU), Hong Kong SAR — joint doctoral programme with GDIIST
Open To: Research internships worldwide, remote-friendly (US time-zone overlap available)