2026 China AI Agent Conference Beijing Stop Announces RSI Workshop Speaker Lineup

On December 15-16, the 2026 China AI Agent Conference (Beijing Stop), initiated by ZhiDongXi and hosted by ZhiXingXing, will officially take place.

As the second thousand-attendee-level AI agent technology and product event held within the year, following the Hangzhou stop, this conference is themed "Paradigm Advance, Innovation Surging" and features five major sections: "Opening Ceremony + Forums + Seminars + Networking Dinner + Exhibition Area," with plans to invite 70+ heavyweight guests to deliver remarks, reports, speeches, dialogues, and discussions.

▲2026 China AI Agent Conference (Beijing Stop) Agenda I. Latest Guest Lineup for the RSI Seminar: Led by ModelBest's Chief Scientist of Intelligent Evolution

The RSI Technology Seminar will be held on the afternoon of December 16 at the main venue. Currently, Li Yuxuan, Chief Scientist of Intelligent Evolution at ModelBest; Hu Mengkang, founder of Evolvent AI; Zheng Yaowei, author of LlamaFactory and PenguinHarness; and Wu Yin, PhD student at The Hong Kong University of Science and Technology (Guangzhou) and research intern at ByteDance's Jindouyun, will attend and deliver keynote reports.

English text is being prepared.

Hu Mengkang is a young entrepreneur and researcher worth watching in the RSI direction. He was selected for the "AI Innovators Under 25" list released by Sequoia China in collaboration with MIT Technology Review China, and led his team to develop the agent project OWL, with related projects cumulatively gaining tens of thousands of GitHub Stars. Hu Mengkang also gave a report at the self-evolving agent seminar at the Hangzhou stop in July, and his current research focus has further extended to recursive self-improvement (RSI).

Zheng Yaowei is a core developer of the previously viral fine-tuning framework LlamaFactory, which has garnered over 70,000 stars on GitHub. The research results have been deployed and applied by companies at home and abroad such as Alibaba Cloud and NVIDIA. He currently serves as CTO of PrismShadow, and has recently open-sourced a new project, PenguinHarness, a Harness project capable of building continuously evolving Agents.

Wu Yin previously led the design of Huawei Cloud's internally built high-precision, low-latency ontology-evolving RSI agent, and also promoted the implementation of RSI in risk control business scenarios at ByteDance. His research mainly focuses on how to enable Recursively Self-Improving (RSI) Agent/RSI for LLM Post-training in enterprise scenarios. Related results have been published at top academic conferences such as AAAI (Oral), NIPS, and EMNLP.

1. Li Yuxuan, Chief Scientist of Intelligent Evolution at ModelBest

Li Yuxuan is the Chief Scientist of Intelligent Evolution at ModelBest, a PhD from the Department of Computer Science at Tsinghua University, and a "Shuimu Scholar" postdoctoral fellow (high-performance computing direction).

He was admitted to Tsinghua University through the Informatics Olympiad, and won five championships in seven domestic and international university supercomputing competitions (including ASC'17 and ISC'17 world championships). He previously served as the core technology lead at the National Supercomputing Center in Wuxi, leading system-level performance optimization of major scientific computing applications on "Sunway TaihuLight," and ran the world's first full-system simulation of 50-photon Gaussian boson sampling; he designed the core simulation algorithm for the "Jiuzhang" photonic quantum computer, helping China achieve "quantum supremacy" for the first time, with results published in the main journal of Science and selected as one of China's Top Ten Scientific Advances of 2020. He once served as the technical lead to build a domestic AI chip autonomous computing power foundation from scratch, building a basic software team of more than 50 people from scratch and coordinating more than 100 people.

English text is being prepared.

He has co-authored more than 40 papers, with nearly 6,000 Google Scholar citations, and has received honors such as the China Youth May Fourth Medal (collective) and the Jiangsu Province "U35 Exploration Award." His research career spans supercomputing, quantum computing, domestic chips, large models, and RSI, making him a composite technical leader connecting "computing hardware" and "AI algorithms."

2. Hu Mengkang, Founder of Evolvent AI

Hu Mengkang graduated with a bachelor's degree from Harbin Institute of Technology and is pursuing a PhD at the University of Hong Kong, researching Embodied AI and LLM-based Agent fields. He is currently starting a business in the LLM-based Agent direction and is the founder of Evolvent AI.

During his doctoral studies, he produced 40+ papers, including 10 first-author/corresponding-author AI papers (ICLR, KDD, NeurIPS, etc.). His Google Scholar citations exceed 2,000. He was selected for Tencent's Qingyun Program, Xiaohongshu's Top Talent Program, and Microsoft Research Asia's Rising Star, and received investment from a top domestic USD fund, among other achievements. He led the team in developing the open-source project OWL, which has accumulated 19,000+ GitHub stars and is one of the fastest-growing projects in stars in the Agent direction. Projects he has personally participated in have received a total of over 30k stars.

Evolvent AI focuses on building Agent Data Infrastructure, with the core vision of creating an AI coworker system capable of continuous learning and self-evolution. Current areas of focus: 1. Agent data & environments annotation; 2. Practical enterprise deployment of self-evolving agents.

His research directions are: (1) a general-purpose AI Assistant Harness for the real world (OWL); (2) Supervised Fine-tuning & Reinforcement Learning for LLM-based Agent (OWL, AgentGen); (3) automated construction of Agent World Model/Environment/Data Pipeline (Text2World, AgentGen, Agent2World, Tool-Genesis); (4) Context Engineering/Optimization for LLM-based Agent (HiAgent, Tree-Planner).

3. Zheng Yaowei, author of LlamaFactory and PenguinHarness

Zheng Yaowei holds a PhD from the Department of Computer Science at Beihang University and is CTO of PrismShadow. He is the author of the LlamaFactory large-model fine-tuning framework, which has earned over 80,000 stars on GitHub. As first author, he has published multiple papers at conferences and journals including ACL, CVPR, and WWW, and serves as a reviewer for AAAI and EMNLP. He has received honors such as Beihang Role Model and the Ascend Ecosystem Open Source Excellence Contribution Award. He is a national first-prize winner of the Huawei Cup National Mathematical Modeling Contest and the Lanqiao Cup Information Technology Competition. He has been invited to deliver keynote speeches multiple times at industry summits such as the Artificial Intelligence Computing Conference and the Alibaba Cloud Apsara Conference. His research results have been deployed and applied by domestic and international companies such as Alibaba Cloud and NVIDIA.

4. Wu Yin, PhD student at The Hong Kong University of Science and Technology (Guangzhou) and ByteDance Jindouyun research intern

Wu Yin is a second-year PhD student at The Hong Kong University of Science and Technology (Guangzhou) and currently a ByteDance Jindouyun intern. He studies under Professor Xiong Hui, a Fellow of AAAI/ACM/IEEE, and Professor Tang Nan, an ACM Distinguished Member. In the open-source community, the Supervisor.skill series of projects he leads has now received over 11k stars on GitHub; his personal GitHub total exceeds 30k stars. His research results have been successfully deployed in ByteDance, Huawei Cloud, and Baidu AI products. His research focuses primarily on Recursively Self-Improving (RSI) Agents/RSI for LLM Post-training in enterprise scenarios. Related results have been published at top academic conferences such as AAAI (Oral), NeurIPS, and EMNLP. II. Progress of the Other Seven Workshops: Tencent Hunyuan, Alibaba Amap, and ByteDance Seed Researchers Gather

Currently, a total of 12 speakers have made their debut. In addition to the RSI Technology Workshop, the other seven technical workshops also bring together guests from the open-source community, universities, and industry.

For the Personal Agent Technology Workshop, we have invited Wu Jiasheng, author of DS Bot (also known as the community edition of DeepSeek Bot).

At the Agentic Video Generation Technology Workshop, Wang Tianfu, an AI PhD student at The Hong Kong University of Science and Technology (Guangzhou), and Wang Weijie, a PhD student at Zhejiang University and ByteDance Seed research intern, will give talks.

The Long-Horizon Agent Technology Workshop has invited Ma Ziyu, a researcher at Alibaba Amap, Zhang Ningyu, an associate professor at Zhejiang University, and Su Qisheng, a joint-training PhD student at USTC & Shanghai Innovation Institute, to give presentations.

Kang Jiazheng, an applied algorithm researcher at Tencent Hunyuan, and Jiao Zhengbo, a PhD student at The Hong Kong University of Science and Technology, will give talks at the Agent Memory Technology Workshop and the Agent Sandbox Technology Workshop, respectively.

More conference guests will be announced as soon as possible.

1. Wu Jiasheng, author of DS Bot

Wu Jiasheng is an independent developer, AI creator, Cognition global ambassador, and a self-media blogger with over 10,000 followers. He focuses on Agent architecture, with particular attention to Agent Memory, Context Management, and multi-Agent collaboration. He is the author of the open-source project DS Bot (a DeepSeek Harness plugin, with nearly 3,000 stars on GitHub), and has also developed the real API price comparison tool RealAPIPricing (nearly 1,000 stars) and the multi-Agent collaboration tool Agent Bridge, with a cumulative total of nearly 4,000 stars on GitHub. He has long shared AI agents, developer tools, and independent development practices on X and video platforms, with monthly views exceeding 7 million.

2. Zhang Ningyu, Associate Professor at Zhejiang University

Zhang Ningyu, PhD, is an associate professor and a Qizhen Outstanding Young Scholar, and has been selected multiple times for Stanford's global top 2% scientists list. His main research directions include natural language processing, large models, knowledge graphs, and knowledge editing. He has led more than ten projects, including National Natural Science Foundation of China Youth/General Programs, a sub-project of the Ministry of Science and Technology's Science and Technology Innovation 2030 Next-Generation Artificial Intelligence Major Project, CCF-Baidu Pinecone/Tencent Rhino-Bird Funds, the CAAI-Huawei MindSpore Fund, and cooperative projects with leading internet companies. He has developed the ocean large model OceanGPT·Cangyuan, the knowledge acquisition system DeepKE, the large-model knowledge editing system EasyEdit, and the large-model lightweight memory system LightMem. He serves as an Associate Editor for Neural Networks, Machine Intelligence Research, ACM Transactions on Asian and Low-Resource Language Information Processing, and IP&M, as an Area Chair for ACL, EMNLP, NeurIPS, ICLR, and KDD, as an ARR Action Editor, as a Senior Program Committee member for IJCAI, and as a reviewer for multiple Nature sub-journals. Students he has co-supervised have been selected for the Doctoral Student Special Program of the China Association for Science and Technology Young Science and Technology Talent Cultivation Project, the China Electronics Society-Tencent Doctoral Student Research Incentive Program, the Chinese Association for Artificial Intelligence Potential Qingyuan Scholars, the Chinese Information Processing Society of China "Doctoral Dissertation Incentive Program," and the Zhejiang Province Qingmiao Talent Program.

3. Kang Jiazheng, Applied Algorithm Researcher at Tencent Hunyuan

Kang Jiazheng is an applied algorithm researcher at Tencent Hunyuan and a master's student in computer technology at Beijing University of Posts and Telecommunications. His main research directions are Agent Memory and large-model post-training. He has published multiple papers at international conferences such as EMNLP, WWW, and AAAI. His representative work Memory OS of AI Agent received an EMNLP 2025 Main Oral, has been cited more than 350 times, and its related open-source project has earned 1.6K stars on GitHub. He is currently involved in the development of Hy-Memory 2.0 at Tencent Hunyuan, responsible for post-training of the memory-specialized model Hy-Memory-30B-A3B and the Coding Memory direction. This model has only about 3B activated parameters, yet its average performance across multiple memory systems surpasses Opus5. Within two weeks of the model's open-source release, the community produced more than 20 quantized and adapted versions, and the entire series has exceeded 110,000 downloads. At this event, he will share practical experience in Agent memory model training and memory system deployment, in the context of Hy-Memory 2.0.

4. Ma Ziyu, Researcher at Alibaba Amap

Ma Ziyu is a researcher at Alibaba Amap, focusing on long-horizon agents and autonomous evolution. As first/co-first author, he has published multiple academic papers at top international conferences such as CVPR, ACL, ECCV, AAAI, and EMNLP; his representative works include TerminalHorizon, LongHorizon-Harness, SkillClaw, and Tree-GRPO. His GitHub projects have accumulated more than 5,000 stars and have attracted widespread attention from researchers at home and abroad.

5. Wang Weijie, PhD student at Zhejiang University and ByteDance Seed research intern

Wang Weijie is a PhD student at the College of Computer Science and Technology at Zhejiang University, studying under Professor Zhuang Bohan. He is currently a ByteDance Seed research intern and previously interned at Microsoft Research Asia. He graduated from Zhejiang University with a bachelor's degree and was selected for the Chu Kochen Honors College Advanced Engineering Education Program. His research directions include world-model agents with 3D perception capabilities, with a focus on real-scene-to-simulation environment construction for robots, scaling of simulation data, and 3D consistency of video world models. Related research results have been published at international conferences such as ICML, NeurIPS, ICCV, and ECCV, with over 1,000 Google citations. He was selected for the Doctoral Student Special Program of the China Association for Science and Technology Young Science and Technology Talent Cultivation Project and has received honors such as the National Scholarship.

6. Wang Tianfu, AI PhD student at The Hong Kong University of Science and Technology (Guangzhou)

Wang Tianfu is an AI PhD student at The Hong Kong University of Science and Technology (Guangzhou), with research focused on the foundations and applications of Agentic AI. He has previously worked on model training and Agent product practice at MSRA, WorkBuddy, and Kimi. He has published more than 30 papers in top AI conferences and journals, has over 1,000 Google Scholar citations, and received an ICML Workshop Best Paper award. He leads open-source projects such as FrontierVideo and MindOS, which have cumulatively received over 2,000 GitHub Stars.

7. Jiao Zhengbo, PhD student at The Hong Kong University of Science and Technology

Jiao Zhengbo is a PhD student at The Hong Kong University of Science and Technology. As first author, he has published multiple papers at international top conferences such as ICML, CVPR, and EMNLP, and has served as a reviewer for top conferences including NeurIPS, ICML, ACL, ECCV, and COLM. He has successively served as a research intern at Alibaba's AIData team (in collaboration with the Qwen team), Tencent Hunyuan's Qingyun Program, and MiniMax Top Talent, participating in the R&D of foundation models such as Qwen 3.5, Hunyuan Hy3, and MiniMax M3.1 Flash. His research focuses on Data-Centric LLM and general Agents, with a core focus on Data Scaling for foundation models, spanning the full stages of Mid-training, Post-training, and Test-Time Training; at present, he focuses on environment augmentation and data synthesis, committed to driving model self-evolution through large-scale synthesis of data and interactive environments, and enhancing models' comprehensive Agentic and reasoning capabilities. Representative works include the Socratic series (Zero, Geo, SWE), Agentic Proposing, SearchEyes, and Policy of Thoughts (selected for EMNLP Oral), among others. Personal homepage: Frostlinx.github.io

8. Su Qisheng, joint-training PhD student at USTC & Shanghai Innovation Institute

Su Qisheng is a joint-training PhD student at USTC & Shanghai Innovation Institute, with main research directions in long-horizon agent evaluation and optimization. Representative works include GraphForge, ACC, Beyond Accuracy, and others. He has participated in and been responsible for optimizing the Search Agent and Working Agent capabilities of multiple cutting-edge industry large models, attracting widespread attention from researchers at home and abroad. III. Exhibition Area Booth Recruitment Officially Launched; First-Day Dinner Also Open for Applications

In the exhibition area of this July's China AI Agent Conference, enterprises such as Tencent Cloud, Zenlayer, X-Chip Technology, Qixin Chenguang, Sangfor Technologies, SenseTime Intelligent, LinkedIn, and NetEase Zhigi made a concentrated appearance, bringing the latest technology and product demonstrations and attracting many attendees for on-site exchanges.

▲Scene of the exhibition area at the 2026 China AI Agent Conference (Hangzhou Station)

In addition, this conference will continue the dinner segment of the previous three conferences this year, holding an exchange dinner on the evening of the first day. The previous three conference dinners cumulatively attracted more than 230 industry leaders, technical experts, and corporate elites. This dinner will build a high-quality exchange and resource-matching platform for attending guests, conducting in-depth discussions around cutting-edge industry trends, technological innovation achievements, and commercial implementation paths, and promoting industrial cooperation and ecosystem co-construction.

▲Scene of the dinner at previous conferences IV. Attendance Channels Now Open

As guest information is gradually disclosed, the attendance channels for this conference have also officially opened. Currently, three types of electronic tickets are available: general pass, VIP ticket, and SVIP ticket.

You can scan the QR code below to add the assistant "Tongtong" for attendance inquiries. Old friends who have already added "Tongtong" can send a private message to "Tongtong" with "Agent" to proceed.

Enterprises with speaking needs, conference sponsorship, or booth sponsorship can also send a private message to "Tongtong" for inquiries.