MarketMind

Inspiration

As a beginner investor, we often hear different financial advice ("All in on NVIDIA, buy high sell low" i. e.) with little to no research and often trade with it. According to SEC, 70% of retail traders experience loses every quarter with most losing 100% in 12 months.

What it does

Introducing, MarketMind, a world simulation of AI agents where each agent emulates a type of trader and allows you to test your investment strategy and how it performs in the market.

How we built it

We built MarketMind with a Python FastAPI backend and React/TypeScript frontend connected via WebSockets for real-time streaming. The core simulation engine uses an order book with price impact and volatility, an orchestrator layer that validates trades and manages portfolios, and LLM-powered agents (via OpenRouter) with distinct trading personalities, quant momentum traders, fundamental value investors, and retail traders. We modeled realistic market dynamics including delayed news propagation (quants react first, retail lags), execution slippage, and market crashes using S&P 500 historical data. A Gemini-powered chatbot provides live market commentary, making the information asymmetry between institutional and retail traders tangible and educational.

Challenges we ran into

Most reliable stock market APIs require payment, and free tiers come with strict rate limits that couldn’t meet our development needs.
Our chatbox initially couldn’t communicate with the server because endpoints weren’t linked correctly.
We hit a lot of bugs under hackathon time pressure, debugging became our secondary hackathon project, and it absolutely tested our emotional resilience.
For some of us, it's also our first hackathon so there was a steep learning curve

Accomplishments that we're proud of

MarketMind is designed to help new traders learn how to “survive” in the stock market through guided, structured learning.
We created our own AI agents and organized them into common real-world types like retail, quant, fundamental, and institutional (and more).
We connected the Google Gemini API to power a chat experience that can interact with users and support their learning journey.

What we learned

We learned how to build a full web app, including front-end, back-end, and making them work together.
We learned methods to shape agent behavior, so they act more consistently and “human-like” for different roles.
We learned key trading/market concepts so our agents and explanations could be more realistic and helpful.
Last, we learned how to divide tasks effectively, and integrate everyone’s work into one working product with collaboration.

What's next for MarketMind

In 3 months: Multiplayer competition mode (hundreds of traders learning and competing).
In 6 months: Real Finnhub market data integration + deeper AI agent reasoning/explanations.
In 1 year: A full teaching tool for finance education (structured lessons + practice + feedback).