← Back to Work

End-to-End Market Simulator

Launch app

Visit on desktop to access

Overview

A multi-agent market simulation with live synthetic news, autonomous trading personas, streaming prices, and a Professor Mode for options concepts.

Project notes

A multi-agent stock-market simulation where autonomous LLM agents trade on synthetic news in real time.

A fictional financial market populated by autonomous trading agents and a narrator model that produces synthetic news. A global narrator drives market sentiment, while three agents react with distinct trading strategies. Michael Chen trades on momentum, Victoria Shaw is a value trader with contrarian tendencies, and Hannah Park shifts focus from position execution to tail-hedging. All trades made by agents feed back into the market state.

The front-end is powered by a React dashboard, which streams prices, headlines, and the agent ledger over WebSockets.

How the simulation is put together

The Python backend separates the market’s world state, the trading agents, and the FastAPI server. The server sends state, news, and trade updates over WebSockets; the React dashboard uses those events to keep the charts and agent activity in step. The market is fictional, so visitors can inspect how each agent reacts without mistaking the output for live trading advice.

As I built this platform while studying these trading strategies, and the stock market as a whole, I implemented a “Professor Mode” which educates on the definitions of the Greeks (Delta, Gamma, Theta, Vega) and the Black–Scholes pricing model.

The simulation only runs while at least one observer is connected, keeping inference cost near zero during development.

The local version can also run without a Gemini API key by using rule-based fallbacks. That made it possible to test the market loop and interface before using model calls for the narrator and agents.