Junjun Liu (Frank)

AI systems & applied machine learning

About

I work on reliable AI systems, distributed training and inference, and tools for machine learning research. I am currently interning at KVCache.AI, contributing to Mooncake.

My experience also includes large language model post-training and fine-tuning. I collaborate with Robert Pollice at the University of Groningen on modeling peptides with AI. My projects span automated paper reproduction, privacy evaluation, gradient monitoring, and machine learning for molecular design.

Experience

KVCache.AI · Mooncake

Current

Intern · AI systems

Contributing to Mooncake's transfer engine, with a focus on RDMA notification reliability and fault tolerance. My work explores endpoint lifecycle handling, control-plane fallback, and notification delivery under failures.

MooncakeNotification fault-tolerance RFC

Robert Pollice Research Group

Current

Research Collaboration · University of Groningen

Collaborating with Robert Pollice on modeling peptides with AI, bringing together machine learning and computational chemistry.

LLM Post-training for AI Podcast Generation

2025

Hunyuan AI Podcast Innovation Challenge · OpenAtom Competition

Worked on post-training and fine-tuning Tencent Hunyuan language models for AI podcast generation.

Competition

FlagScale

2026

Software Engineering Intern · Remote

Worked on fault tolerance and observability for distributed training, including straggler detection, in-process restart support, and structured training-health monitoring.

MetaX

2026

Project Contributor · Remote

Worked on AI compiler subgraph optimization, including pattern matching, graph rewriting, backend-compatible execution, and correctness and latency profiling.

Shanghai Media Group

Jun–Aug 2022

Machine Learning Engineer · IT Department

Built real-time models for interactive virtual environments, combining lightweight visual feature extraction, temporal gesture recognition, and multi-person detection.

Research & Projects

Automated Research-Paper Reproduction

May 2026

Bohrium “Light Chaser” Paper-Reproduction Sprint · 2nd place

Built an agent harness connecting PDF parsing, method extraction, code generation, environment setup, execution, and iterative debugging. Compared reproduced metrics with reported results and generated structured reproduction reports.

ColossalAI Monitor

Mar–May 2025

Team Leader · Distributed gradient monitoring

Designed non-intrusive PyTorch autograd hooks to capture per-parameter gradient statistics in ColossalAI and PyTorch DDP. Compared local and global gradients around AllReduce and exported structured records for analysis.

Inverse Design of Cell-Penetrating Amphiphiles

Oct 2024–Apr 2025

Robert Pollice Research Group · University of Groningen

Built data and modeling pipelines for gene-delivery amphiphiles, combining molecular representations, RDKit features, and multi-task neural networks. Trained and evaluated PyTorch models on HPC GPUs.

AI Application Deployment on openEuler

Jan–Apr 2024

Project Leader · openEuler Community / OpenAtom Foundation

Packaged PyTorch and its ecosystem as reproducible RPM builds on openEuler. Developed an offline deployment pipeline and Gradio demonstrations for speech synthesis, sentiment analysis, image classification, and image generation.

Efficient Fine-Tuning & Prompt Engineering

Aug–Oct 2023

LoRA · Ascend 910 · Generative AI

Created custom image datasets and explored LoRA-based adaptation on Ascend hardware, combining parameter-efficient fine-tuning with prompt engineering for controlled generation.

Publications

Investigating Expert-Based Prompt Engineering for Legal Entailment Tasks

Cor Steging, Ludi van Leeuwen, Tadeusz Zbiegień, Dries Wedda, Junjun Liu

COLIEE 2025 Workshop · Chicago, USA · June 2025

Paper (PDF)Workshop

Education

Northeastern University

2026–2028 · Expected

M.S. in Artificial Intelligence · Incoming

Khoury College of Computer Sciences · Machine Learning Concentration

University of Groningen

Graduated August 2026

B.Sc. in Artificial Intelligence

Coursework in machine learning, NLP, algorithms, mathematics, statistics, and AI systems.

Honors & Awards

Mathematical Modeling