SHANGHAI — In a move that has stunned the global tech community, Guo Hangjiang, a 20-year-old senior at Beijing University of Posts and Telecommunications, has transformed a 10-day “vibe coding” project into a multimillion-dollar startup. His project, MiroFish, an open-source multi-agent simulation engine, skyrocketed to the #1 trending spot on GitHub earlier this month, amassing over 42,000 stars. The momentum culminated in a 30 million RMB (~$4.1 million USD) seed investment from Shanda Group founder Chen Tianqiao, finalized just 24 hours after a viral demonstration.
MiroFish is designed to simulate “parallel digital worlds” by populating them with thousands of autonomous AI agents. By ingesting real-world data—ranging from policy drafts to financial reports—the system allows users to “stress test” reality, observing how digital societies react to specific events before they happen in the real world.
Beyond the Hype: How MiroFish Actually Works
While social media clips often frame MiroFish as an “AI crystal ball,” the technical reality is more grounded. The system utilizes a sophisticated GraphRAG (Graph Retrieval-Augmented Generation) architecture to build its world model.
- Knowledge Graph Construction: Unlike traditional RAG, which treats data as flat text, MiroFish builds a structured network of entities (people, institutions, events) and their relationships.
- Agent Generation: The engine spawns thousands of agents, each assigned a unique personality, memory, and behavioral logic derived from the graph.
- The OASIS Simulation: Using the OASIS framework (developed by CAMEL-AI), these agents interact on parallel platforms resembling social media environments, allowing for emergent behaviors like “herd effects” and “opinion polarization.”
From “Dream of the Red Chamber” to Global Markets
MiroFish has already demonstrated its versatility through high-profile demos. In one instance, the team fed the first 80 chapters of the classic Chinese novel Dream of the Red Chamber into the engine to simulate and “predict” its lost ending based on character dynamics.
More practical applications are gaining traction in the corporate sector. Firms are reportedly using MiroFish to simulate public opinion crises and predict market reactions to supply chain shocks. The project’s ability to provide a “God’s-eye view” of a simulated society offers a level of interpretability that traditional “black-box” machine learning models cannot match.
Expert Analysis: The Rise of the “Super-Individual”
The success of MiroFish signals a tectonic shift in the AI industry: the era of the Super-Individual. Guo’s ability to build a world-class simulation engine in 10 days—previously a task requiring dozens of engineers—proves that AI-assisted development has lowered the barrier to entry for complex infrastructure.
“Chen Tianqiao isn’t just betting on the software; he’s betting on the collapse of traditional R&D timelines,” says an industry analyst. “MiroFish represents ‘Grounded Emergence.’ It’s the first time we’ve seen multi-agent simulations move from academic toys to decision-support tools that are actually accessible to the open-source community.”
However, experts caution against over-reliance on these simulations. Because MiroFish relies on LLM reasoning, it is susceptible to consensus collapse and RLHF bias, where agents may lean toward “polite” or “safe” outcomes rather than the messy, unpredictable nature of real human behavior. The long-term impact of MiroFish will likely not be in “telling the future,” but in helping leaders identify “blind spots” in their strategic planning.
READ: Claude Data Extraction Controversy Reshapes Global AI Competition
5 FAQs About MiroFish
1. Can MiroFish actually predict the future?
No. It performs “scenario simulation.” It shows plausible outcomes based on current data and agent logic, acting as a stress test rather than a guaranteed forecast.
2. Who is the creator of MiroFish?
Guo Hangjiang (also known as “Baifu”), a 20-year-old undergraduate student in China who previously created the viral sentiment analysis tool BettaFish.
3. What is the “Vibe Coding” mentioned in reports?
It refers to a rapid, intuitive development style where the coder uses AI agents and LLMs to handle the bulk of the boilerplate, focusing only on the high-level architecture and “vibe” of the product.
4. Is MiroFish free to use?
Yes, it is open-source under the AGPL-3.0 license. You can find the code on GitHub under the repository 666ghj/MiroFish.
5. How much does it cost to run a simulation?
Costs can be high because simulating thousands of agents requires significant LLM API calls. However, a “MiroFish-Offline” fork exists for users with powerful local GPUs to run simulations without API fees.
