# That's Eye-Opening — A Undergrad in Quantitative Finance Built an AI Cluster Engine in 10 Days, and It's Blowing Up on GitHub with 13,000+ Stars and $4 Million in Funding

There's a project called MiroFish that's been trending lately. It's basically a multi-agent simulator designed to help you make predictions — whether it's trading, PR, or anything else.

## What Exactly Is MiroFish?

Simply put, it's a digital "sandbox world." Thousands of AI agents run inside it, each with their own "memories" and behavioral patterns, interacting like real people — discussing, debating, taking sides. You throw any scenario into it — say, a piece of news suddenly leaks, policies change, or even an unfinished classic novel — and it simulates how these agents would react, argue, and what outcome they'd reach, helping you predict what might happen in reality.

The person behind this project is Guo Hanjiang. He was still an undergrad in late 2025. Back then, he used AI tools to help write code, got the core functionality working, and when he uploaded it, it exploded. The Shanda Group immediately invested 30 million RMB. Later, he moved out of his dorm and started his own company, now pushing this direction forward.

## What Can You Use It For?

Currently, the most popular use cases include:

**Trading:** Drop financial news or earnings reports into it, and it simulates whether the market will panic, how prices might move — giving you a predictive reference.

**PR Testing:** A company or team drafts a statement, runs it through first to see if it'll backfire, then adjusts the wording accordingly.

**For Fun:** Some people put an unfinished Chinese novel into it, and the agents "play" the characters, working out a logical ending themselves.

Deployment is simple — Docker gets it running in minutes, and it comes with API keys to connect to various large language models.

## Pro Tips for You

You can even pull off some "bold moves" — like simulating Elon Musk suddenly tweeting "Dogecoin 2.0 is here," then watching traders, influencers, and retail investors go crazy inside the simulation. The whole process can even generate video, giving you a completely risk-free taste of what "moonshot" or "crash" feels like.

Traders have already made money with this. For example, there's a case on Polymarket where someone ran MiroFish on historical data, simulated the scenario, then bet on SPX 500 and made over $120,000 net profit.

His address is:

If you want to take a shortcut and just follow the pros, you can also try Kreo's copy trading — automatically replicate operations from people like him, ride the wave.

Add his wallet [0x17559efac103ac7f361be37ec0b93888d4c55aac] on PolyCop to start tracking or copying his trades.

Code is here: