Inspiration
Traditional algorithmic trading lacks the nuanced reasoning that human traders bring. We wanted to create a system that combines the speed of automation with the strategic thinking of experienced traders, using multiple AI agents that collaborate like a real trading desk.
What it does
Narrative Alpha orchestrates five specialized AI agents:
- Trading Director - Generates trading theses and makes final decisions
- Quantitative Analyst - Analyzes technical indicators (RSI, MACD, Bollinger Bands)
- Risk Manager - Calculates position sizing using Kelly Criterion
- Execution Agent - Generates structured trade orders
- Sentiment Agent - Monitors news and social media sentiment
Together, they analyze market data, detect sentiment-price divergences, and produce actionable trade recommendations.
How we built it
We used the Swarms framework to orchestrate multiple AI agents powered by Anthropic and OpenAI models. The backend runs on FastAPI with PostgreSQL and Redis. Data flows from multiple sources (yfinance, Finnhub, NewsAPI, Twitter) through our collectors, gets processed by the agent hierarchy, and outputs JSON trade recommendations. A React/TypeScript dashboard visualizes everything.
What we learned
Multi-agent systems require careful prompt engineering for consistent collaboration Sentiment analysis adds significant alpha when detecting market divergences The Kelly Criterion needs adjustment for volatile assets Human-in-the-loop validation remains crucial for high-stakes decisions
Built With
- anthropic
- docker
- fastapi
- finnhub
- framework
- gpt
- newsapi
- openai
- postgresql
- python
- react
- typescript
Log in or sign up for Devpost to join the conversation.