Mobius

Inspiration

We noticed a fundamental gap in the trading world: powerful algorithmic strategies were locked behind coding barriers. Retail traders had ideas—"buy when Elon tweets positively about Tesla" or "trade based on Fed sentiment"—but lacked the technical skills to execute them. Meanwhile, institutional traders spent thousands on custom bots that took weeks to develop. We asked: what if AI could bridge this gap? Mobius was born from this vision. We wanted to democratize algorithmic trading by letting anyone describe their strategy in plain English. Our multi-agent AI system handles the complexity—analyzing sentiment, backtesting with real historical data, and generating production-ready code. No Python knowledge needed. No expensive developers required. We're making sophisticated trading accessible to everyone, turning market intuition into automated action.

What it does

Mobius transforms natural language trading ideas into fully functional, backtested trading bots. Users describe their strategy in plain English—like "trade Tesla based on news sentiment and RSI indicators"—and our platform handles the rest. Our multi-agent AI system orchestrates the entire workflow: one agent analyzes the request and selects appropriate technical indicators, another gathers real historical market data and sentiment from sources like Finnhub and Reddit, while a third agent writes production-ready Python code. The system then backtests the strategy against historical data, showing actual performance metrics like returns, win rate, and drawdown. Users receive a complete trading bot they can download and deploy immediately. The platform visualizes backtest results with interactive charts, displays generated code (with download options), and manages a library of saved strategies. Everything happens automatically—no coding, no manual data gathering, no guesswork about whether a strategy actually works.

How we built it

Mobius is powered by a React + Vite frontend and FastAPI backend, orchestrated through a multi-agent AI architecture using Google's Gemini API. The frontend provides an intuitive interface where users describe their trading strategies. We integrated Supabase for authentication and database management, storing user strategies and backtest results. The UI features real-time progress tracking through Server-Sent Events (SSE), showing users exactly what each AI agent is doing. The backend coordinates three specialized AI agents: a strategy analyzer that interprets user intent and selects technical indicators, a data agent that fetches real historical market data from Alpaca and Finnhub APIs, and a code generator that writes production-ready Python trading bots. We implemented robust backtesting using real historical price data and sentiment analysis from multiple sources including Finnhub news and Reddit (via PRAW). For technical analysis, we integrated TA-Lib and pandas-ta libraries. The system validates strategies through actual backtesting—not simulations—using historical data going back months or years, providing users with accurate performance metrics before they risk real capital.

Challenges we ran into

he backend presented our toughest obstacles. Initially, our bot generation endpoint ran synchronously, causing timeouts when AI agents took too long to analyze strategies or fetch historical data. We had to architect an asynchronous task system with result caching and Server-Sent Events for real-time progress updates. Backtesting proved incredibly complex. Early versions crashed with cryptic NoneType errors when sentiment data was missing for certain dates. We built robust fallback mechanisms—trying Alpha Vantage first, then Finnhub, finally graceful degradation to mock data when APIs hit rate limits. Coordinating multiple data sources (Alpaca for prices, Finnhub for news, Reddit for social sentiment) while respecting rate limits required careful caching and request throttling. The multi-agent orchestration was particularly tricky. Getting three AI agents to work together—passing context between strategy analysis, data gathering, and code generation—required extensive prompt engineering. We burned through API quotas debugging edge cases where agents misinterpreted user intent or generated invalid Python code. Twitter API costs ($100/month for basic access) forced us to pivot entirely to Finnhub and Reddit for sentiment analysis, requiring significant rewrites of our data pipeline.

Accomplishments that we're proud of

We're incredibly proud of building a genuinely functional multi-agent AI system that actually works. This isn't a demo with hardcoded responses—Mobius generates real, executable trading bots that users can download and run immediately. Our backtesting engine uses real historical data, not simulations. When users see performance metrics, they're based on actual market data from Alpaca and Finnhub, with real sentiment analysis from Reddit and news sources. The accuracy gives users confidence before risking capital. The seamless user experience is another major win. Users simply describe their idea in plain English, and within minutes receive a complete trading bot with backtested results, performance charts, and downloadable Python code. No coding knowledge required. We successfully integrated multiple complex APIs (Gemini for AI, Alpaca for trading data, Finnhub for news sentiment, Supabase for auth) into a cohesive system that feels simple to use. The real-time progress tracking through SSE keeps users engaged, showing exactly what each AI agent is doing behind the scenes. Most importantly, we democratized algo trading—making institutional-grade strategies accessible to anyone with a trading idea.

What we learned

We learned that AI agent orchestration is far more nuanced than simply chaining API calls. Effective prompt engineering was critical—each agent needed precise instructions, context about what previous agents discovered, and clear output formats to pass data seamlessly. We iterated dozens of times to get agents communicating properly. API rate limits and costs became a crash course in resource management. Twitter's $100/month pricing taught us to evaluate data sources not just on quality, but on accessibility. We learned to build robust fallback systems—when one API fails or hits limits, gracefully degrade to alternatives without breaking the user experience. Backtesting with real historical data revealed the gap between theory and practice. Our initial naive implementations crashed on missing data, timezone mismatches, and edge cases we never anticipated. We learned that production-grade financial software requires obsessive error handling and data validation. On the frontend, we discovered the power of real-time feedback. Users stayed engaged when they could see AI agents working, transforming what felt like a slow process into an exciting experience. Most importantly: democratizing complex technology requires hiding complexity, not dumbing it down.

What's next for Mobius

We're planning to expand Mobius into a comprehensive trading platform. First priority is adding live paper trading—letting users test their bots in real-time with simulated funds before risking real capital. This bridges the gap between backtesting and live trading. We want to build a community marketplace where users can share successful strategies, rate bots, and learn from each other's approaches. Imagine browsing top-performing sentiment-based strategies or discovering new indicator combinations that actually work. Enhanced data sources are next—integrating more sentiment APIs like StockTwits, expanding to cryptocurrency markets, and adding alternative data like options flow and institutional holdings. We're also exploring real Twitter API integration for users willing to pay for premium data. On the AI side, we're developing agents that can optimize existing strategies, suggest improvements based on backtest results, and even propose entirely new trading ideas based on market conditions. Finally, we're building portfolio management features—letting users run multiple bots simultaneously with intelligent position sizing and risk management across their entire trading account. The goal: make Mobius the go-to platform for AI-powered retail trading.