·
Builders & Technical Leaders
Sirui Hong
Co-Founder & Technical Lead |DeepWisdom
Creator of MetaGPT, the first multi-agent framework to simulate an AI software company with SOPs. ICLR 2024 Oral (#1 LLM-Agent). Also first-authored Data Interpreter (ACL 2024) and co-authored AFlow (ICLR 2025 Oral). Leads DeepWisdom's NLP/AIGC algorithms; launched MGX/Atoms reaching 500k users in one month.
3papers
GitHubgeekanResearch Papers
3 papersBiography
Deep Research
Sirui Hong is the co-founder of DeepWisdom and the technical leader of NLP/AIGC Algorithms at the company, best known as the creator and first author of MetaGPT — the pioneering multi-agent framework that models an entire AI software company with product managers, architects, project managers, and engineers collaborating via Standardized Operating Procedures (SOPs). The MetaGPT paper received an oral presentation (top 1.2%) at ICLR 2024, ranking #1 in the LLM-based Agent category. Hong is also the first author of the Data Interpreter paper (ACL 2024) and a co-author of AFlow (ICLR 2025 Oral, top 1.8%). Under the FoundationAgents GitHub organization, MetaGPT has 65k+ stars and OpenManus has 55k+ stars. Hong's research focuses on multi-agent collaboration, automated workflow generation, and code generation. DeepWisdom, a Shenzhen-based company founded in 2019, launched MGX (MetaGPT X) in February 2025 — later rebranded as Atoms in January 2026 — reaching 500k users in its first month and crossing $1M ARR. In the first half of 2025, DeepWisdom raised ~$30.8M (RMB 220M) from Ant Group, Cathay Capital, Baidu Ventures, and others, representing the largest financing in the domestic coding agent track.
multi-agent collaborative frameworksmeta programming for LLM agentsAI software company simulationautomated agentic workflow generationLLM-based code generationstandardized operating procedures for AI agentsdata science automation with LLM agentsvibe coding and AI-powered development
Timeline
14 Research14 total
2025
2024
2023
Key Contributions
MetaGPT Framework
Created MetaGPT, the first multi-agent framework to model an entire AI software company with distinct roles (product manager, architect, project manager, engineer) collaborating via Standardized Operating Procedures. Encodes SOPs into prompt sequences for structured multi-agent collaboration, achieving state-of-the-art code generation results. ICLR 2024 Oral (top 1.2%, #1 in LLM-based Agent category). 65k+ GitHub stars, 1,489 citations.
Data Interpreter
Developed Data Interpreter, an LLM agent for automated data science that uses hierarchical graph modeling to decompose complex problems into manageable subproblems and programmable node generation to iteratively refine code. Accepted at ACL 2024 with 182 citations.
AFlow: Automating Agentic Workflow Generation
Co-authored AFlow, a system for automatically generating agentic workflows that outperforms all manually designed methods by 5.7% on average and surpasses contemporary automatic workflow optimization by 19.5%. ICLR 2025 Oral (top 1.8%, #2 in LLM-based Agent category). 169 citations.
OpenManus
Core contributor to OpenManus, an open-source general AI agent framework under the FoundationAgents organization enabling users to complete complex tasks (coding, information retrieval, file processing, web browsing) through natural language instructions. 55k+ GitHub stars.
MGX / Atoms Platform
Technical leader behind MGX (MetaGPT X), the world's first AI multi-agent development team product launched February 2025. Rebranded as Atoms in January 2026 with production-ready features including databases, auth, deployment, and payments. 500k users in first month, $1M+ ARR.
Notable Quotes
“
It is true that the success rate of solving many problems increases as the capabilities of larger models increase, but the problems themselves do not go away.
“
There are still a large number of extremely complex problems with long-tail effects in human society, including machine learning, code fixing, and problems that require searching for combinations.
“
The most important thing is to maximize the tasks and effects in real scenarios, including personalization features.
“
The business prospects of multi-agent systems are clear and strong in terms of really effectively solving the actual needs of users, and code generation is a scenario that Agent technology is currently able to solve better.
12 sources(click to expand)
MetaGPT GitHub Repository (FoundationAgents)MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework (arXiv)ICLR 2024 Oral — MetaGPTFrom MetaGPT to Atoms: DeepWisdom leads China's push into vibe coding (KrASIA)China's Coding Agent Secures Largest Financing (36Kr)Sirui Hong — GOSIM AI Paris 2025 Speaker ProfileOpenManus Core Author Interview (KDJingPai)Sirui Hong — Semantic Scholar ProfileOpenManus GitHub RepositoryMGX (Now Atoms) — Product HuntDeepWisdom Official WebsiteWinning solutions of the ChaLearn AutoDL challenge 2019 (arXiv)
Research generated March 19, 2026
Builders & Technical Leaders/Sirui Hong
All Profiles