MIROFISH TOPPED GITHUB TRENDING AND GOT SHANDA MONEY TO SIMULATE THE FUTURE

## An Undergraduate, a GitHub Repo, and a Billionaire's Check MiroFish hit the top of GitHub's Global Trending list in March 2026, an open source AI prediction engine built primarily by undergraduate student Guo Hangjiang. Shanda Group founder Chen Tianqiao, whose net worth Bloomberg pegs at approximately $3.6 billion, invested in the project shortly after. The product does something specific: it ingests real world signals, news articles, financial reports, policy drafts, and constructs a simulated world populated by thousands of AI agents. Those agents interact, argue, trade, and form opinions. Their collective behavior produces prediction reports about how situations might unfold. The architecture has three layers. First, GraphRAG (Graph based Retrieval Augmented Generation) extracts entities and relationships from seed material and builds a structured knowledge graph. That graph becomes the simulated world's reality. Second, MiroFish generates AI agents, each with distinct personalities, memory systems, and behavioral logic. Third, those agents are dropped into a dual platform environment mimicking social networks like Twitter and Reddit, powered by the OASIS engine from the CAMEL AI research community. The agents post, reply, form factions, and shift opinions. MiroFish reads the patterns. ## What It Actually Does A financial analyst feeds MiroFish a set of earnings reports and macro indicators. The system creates a simulated trading floor populated by agents coded with different risk tolerances, investment horizons, and information processing speeds. Some agents panic sell on negative headlines. Others buy the dip. The simulation runs forward, and MiroFish generates a scenario report showing probable price movements, sentiment shifts, and cascade risks. Not a single prediction. A distribution of outcomes. Policy researchers can test draft legislation the same way. Feed MiroFish a proposed regulation, and it simulates how different stakeholder groups, industry lobbyists, consumer advocates, media commentators, engage with the policy over weeks. The output is a forecast of public discourse, not the policy outcome itself but the conversation that shapes the outcome. The tool has also been used for creative applications. Users have fed novel manuscripts into MiroFish to generate plausible plot endings, and media researchers have simulated how misinformation might spread through a social network. ## The Limitations Are the Story MiroFish version 0.1.0 shipped in December 2025. It is early software with real constraints. Running a comprehensive simulation requires thousands of LLM API calls, which means cost scales aggressively with scenario complexity. A single run with 5,000 agents simulating 30 days of interaction could cost hundreds of dollars in API fees depending on the model provider. That makes it a research tool, not a consumer product. The agents also exhibit herd behavior more readily than real humans. When a dominant narrative emerges in the simulation, agents tend to pile on rather than maintain heterogeneous positions. Real social systems have friction, stubbornness, irrationality, and institutional resistance that LLM agents model imperfectly. MiroFish acknowledges this in its documentation, which describes the tool as producing plausible scenarios rather than definitive predictions. ## Why It Matters The underlying idea, multi agent simulation for scenario planning, is not new. The RAND Corporation ran wargame simulations in the 1950s. What is new is the cost curve. MiroFish can spin up a thousand agent simulation in minutes using commodity LLM APIs, a process that previously required custom software, dedicated servers, and PhD level expertise. Open sourcing the engine means a graduate student at any university can run policy simulations that five years ago required a defense contractor's budget. Chen Tianqiao's investment signals institutional confidence in the multi agent simulation category. Shanda Group sits at the intersection of gaming, AI, and data infrastructure, the same territory MiroFish occupies. The GitHub trending position matters because it indicates organic developer interest, not manufactured hype. At version 0.1.0, MiroFish is a research prototype with a billionaire backer and a growing open source community. Whether it becomes infrastructure or remains an experiment depends on whether the cost per simulation drops faster than the complexity of the questions people want to ask it.