Skip to content

0010SS/symmetry

Repository files navigation

ChatGPT Image Sep 14, 2025, 03_02_46 AM

Symmetry: Agentic AI Processing Real-Time Market Info

You know that moment in conversation—
"Man, I wish I bought Tesla back in 2019..."
"She made how much just by following the news early?"

It always seems like someone else saw the wave coming… and rode it.

Is keeping up with the market — the edge of foresight — only for Wall Street banks and billion-dollar hedge funds?

We don’t think so.

🧠 Inspiration

We asked a simple question:
Why can't an individual have access to the same informational edge as a firm with a floor of analysts?

With the explosion of public data and the rise of intelligent AI agents, we realized something:
You can now run your own AI-powered research firm — from your laptop.

This is why we named ourselves, "Symmetry". Symmetrical information for everyone's access.

🤖 What We Built

We designed and deployed a multi-agent AI trading research platform, inspired by real-world investment desks, but driven entirely by 10+ autonomous AI analysts.

We integrated SOTA open-source trading agents and then leveled them up with (i) 2x larger new analysis pipelines, (ii) more in-depth debate modules, (iii) more strategic information ingestion, and (iv) more versatile output.
Now, they read the news, analyze fundamentals, debate risks, and even make strategic recommendations — all in sync.

(It’s like having your own Goldman Sachs... except it’s just you, a keyboard, and some very smart code.)

⚙️ Under the Hood

WechatIMG1532

Our system works like a mini Wall Street firm in a box — with AI agents taking on each role in the investment pipeline:

1. Company-Centric Analysts

  • Stock Trend Analyst → Spots indicators, patterns, and seasonality.
  • Social Media Analyst → Tracks sentiment and headlines from the crowd.
  • Public News Analyst → Monitors macroeconomic news and competitor signals.
  • Fundamentals Analyst → Digs into financial statements and ownership structures.

2. Industry-Centric Analysts

  • Cross-Signals Analyst → Maps supply chain propagation and benchmarks companies against their industry peers.
  • Industry Trend Analyst → Identifies the industry to which the company belongs, spots indicators and patterns.
  • Industry Fundamentals Analyst → Digs into financial statements and ownership structures of the industry's major representative firms.
  • Industry Social Analyst → Tracks sentiment and headlines of the entire industry.

3. Debate Engine

  • Bullish Researcher → Argues for a buy.
  • Bearish Researcher → Argues for a sell.
  • Neutral Judge → Listens to both sides and scores the debate.

4. Trader Agent

  • Decides: Buy, Sell, or Hold?

5. Risk Managers

Evaluates the trade through three perspectives:

  • 🚀 Aggressive Manager
  • ⚖️ Neutral Manager
  • 🛡️ Conservative Manager

6. Fund Manager

Synthesizes everything into a strategy report, covering ideas about:

  • 📅 Annual Investments → grounded in fundamentals & industry growth
  • 📈 Swing Trades → capturing weekly to monthly momentum
  • Intraday Trades → exploiting daily volatility

The result:
A self-contained AI analyst firm that debates, trades, and manages risk — without a single human lifting a finger. A visual pipeline can be found in the image attached.

💡 What We Learned

  • Autonomy requires clarity: To make agents think, we had to teach them to talk. Prompt engineering wasn’t a one-liner — it was diplomacy.
  • Markets are noisy: Parsing real-time sentiment and fundamentals from structured and unstructured data challenged everything we knew about data flows.
  • Modularity matters: A true "firm of one" has to scale — and break — elegantly. We learned to design for chaos.

🚧 Challenges We Faced

  • Aligning multi-agent debate without logic loops
  • Building a coherent world model from fragmented market data
  • Designing human-style strategy prompts that agents understood
  • Keeping the system lightweight, modular, and easy to fork
  • Handling different data structures from APIs (and some were deprecated)
  • Connecting backend output (free-form LLM-generated text) to frontend (requiring structured formats to be inserted into respective fields)

🚀 The Vision

What if…
You didn’t need a Bloomberg terminal or a million-dollar research team to stay ahead of the curve?

What if you could run your own AI-powered investment desk?
One that learns. One that debates. One that sees what others miss.

This isn’t a dream.
It’s our project.
And it’s just getting started.

📎 Try It Yourself

Source Code: https://github.com/0010SS/symmetry
Join us. Democratize information.

Important Note: our codebase structure is adapted from the amazing TradingAgents repo with Apache license. We've added custom prompts for all agents and implemented an extensive set of new agent nodes, tools, and edges. We sincerely appreciate their solid work and code.

Yours truly, Zehao Wen & Zihuan Zhang

About

We are making the market information symmetrical

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 14