Yongfeng Gu

Agentic AI Engineer

I'm dedicated to making AI remember, work, and grow, like we do. Along the way, I've worked across agent harnesses, agent research, and post-training infrastructure, combining research with systems engineering, with a growing focus on how models learn and evolve. With a background in computer vision, full-stack, and embedded development, I am also interested in embodied AI, spatial intelligence, and world models.

I'm an M.S. student at Tianjin University (2024–2027), advised by Prof. Yongjie Ren. Research on vision-based measurement. Seeking 2027 new-grad opportunities.

News

2026.09.07

News Joined MindLab as a Research Intern, building mint-recursive, the infrastructure for model RSI.

2026.07.24

Talk Gave an invited talk on Foundation Protocol at the DataFun Agentic AI conference.

2026.06.16

Patent Laser reference-line positioning system project granted 6 patents (granted June 16).

2026.05.22

Paper Foundation Protocol preprint released — a graph-native coordination layer for the agentic society.

2026.05.13

Paper Harnessing Agentic Evolution (AEvo) submitted to NeurIPS 2026.

2026.03

News Joined DeepWisdom as a Research Intern, working on Agent-related research.

2026.01.21

Paper Co-authored paper published in Chemical Science (Royal Society of Chemistry).

Publications

Harnessing Agentic Evolution

Jiayi Zhang, Yongfeng Gu, Jianhao Ruan, Maojia Song, Yiran Peng, Zhiguang Han, Jinyu Xiang,..., Bang Liu, Chenglin Wu, Yuyu Luo

Reframes agentic evolution as an interactive environment: instead of generating solutions directly, a meta-agent edits the workflow / task-agent that controls later evolution, giving a unified interface for both workflow- and agent-based evolution over long horizons. It outperforms strong baselines on ARC-AGI-2 and Terminal-Bench.

Foundation Protocol: A Coordination Layer for Agentic Society

Bang Liu, Yongfeng Gu, Jiayi Zhang, Zhaoyang Yu, Sirui Hong, Maojia Song, Xiaoqiang Wang,..., Qiang Yang, Yuyu Luo, Chenglin Wu

A graph-native coordination layer for the AI society. It models heterogeneous entities — agents, tools, resources, people, and institutions — as addressable nodes, and organizes coordination across a four-plane architecture: Entity & Trust, Transport & Routing, Interaction & Organization, and Regulation & Oversight.

Organic Neuromorphic Vision Devices with Multilevel Memory for Palmprint Identification

Chenxi Liu, Yongfeng Gu, Yongjie Ren, Mingyi Ding, Tao Deng, Harald Fuchs

Built a five-class palmprint recognition pipeline from 850 augmented CASIA samples: 240×240 ROI preprocessing, CLAHE, block-wise optoelectronic feature mapping, and 36-dimensional feature extraction. Trained GBDT and XGBoost classifiers, reaching 99.02% peak accuracy.

Experience

2026.09 - Present

Research Intern, MindLab, Shenzhen, China.

2026.03 - 2026.08

Research Intern, DeepWisdom, Shenzhen, China.

2025.07 - 2026.02

Engineering Intern, Karis, Beijing, China.

Selected Projects

Foundation Protocol : Multi-Agent Collaboration Runtime

Core Contributor · FoundationAgents

2026.03 - 2026.08

A Python runtime for multi-entity AI collaboration — agents, humans, and tools working together on a shared protocol layer.

AI Link Net : AI Team Collaboration Framework

Core Contributor · FoundationAgents

2026.03 - 2026.08

Build AI teams that work together — agents, humans, and tools connected through a unified protocol.

DCPAM : Laser Reference Line Point-Positioning System

Algorithm Engineer · China Energy Engineering Group

2025.08 - 2026.12

Laser-based shaft alignment measurement for large rotating equipment, achieving 0.02 mm engineering accuracy. First-generation device delivered and in field use; 6 patents granted and 1 paper submitted to a Chinese core journal.

Huixue Education System

Full-Stack Developer · Huixue Education

2024.12 - 2025.02

Built an order management system deployed across 6 cities, serving 2,000+ daily active users and handling 300+ orders per day.

Deep Learning-Based Dosage Prediction for Pediatric Pneumonia

Algorithm Engineer · Tianjin Children's Hospital

2024.07 - 2024.09

Trained CNN models to assess pneumonia severity from chest X-rays, and built an inference application that provides RMPP treatment dosage recommendations for clinicians.