AlphaCouncil: Autonomous Investment Committee 🏛️

AlphaCouncil is an agentic risk-overlay system that orchestrates a "Man-vs-Machine" debate to validate algorithmic trading signals.

Unlike traditional black-box quant models, AlphaCouncil uses a Multi-Agent Architecture (powered by LangGraph) to simulate a hedge fund Investment Committee. It combines deep-learning volatility forecasts with semantic reasoning to filter out false positives caused by event risk (earnings, macro news) or sector concentration.

🏗 Architecture

AlphaCouncil operates as a Directed Acyclic Graph (DAG) with three specialized agents:

1. The Technician (Quant Agent):

   • Role: Signal detection & Regime classification.
   • Core Engine: VolSense (Custom PyTorch Volatility Forecaster).
   • Logic: Analyzes Term Structure, Z-Scores, and Volatility Cones.

2. The Fundamentalist (Research Agent):

   • Role: Event Risk & Sentiment analysis.
   • Tools: Tavily Search API / RAG.
   • Logic: Scans for earnings calls, lawsuits, and macro headwinds to reject "gambling" setups.

3. The Risk Manager (Risk and Execution Agent):

   • Role: Portfolio construction & Limits.
   • Logic: Enforces Sector Limits, Correlation checks, and CVaR constraints.

🚀 Quick Start

Prerequisites

• Python 3.10+
• OpenAI API Key (or Anthropic/Gemini)
• Tavily API Key (for web search)

Installation

git clone [https://github.com/rahulmkarthik/AlphaCouncil.git](https://github.com/rahulmkarthik/AlphaCouncil.git)
cd AlphaCouncil
pip install -r requirements.txt

Usage

Run the dashboard to see the agents in action:

streamlit run app/Home.py

🧠 Core Technologies

• VolSense: Custom Deep Learning Library for Volatility Forecasting (LSTM/Transformer).

• LangGraph: Stateful multi-agent orchestration.

• Streamlit: Interactive frontend for signal visualization.