Financial Intelligence System

A multi-agent financial analysis system that provides real-time stock analysis and trading recommendations for NYSE-listed stocks. The system uses LangGraph to orchestrate a pipeline of specialized agents that perform data ingestion, technical analysis, risk assessment, and AI-powered synthesis.

Features

• Data Ingestion: Fetches 5 years of historical stock data using yfinance
• Technical Analysis: Calculates RSI (Relative Strength Index) and MACD (Moving Average Convergence Divergence) indicators
• Fundamental Analysis: Retrieves key metrics including P/E ratio, market cap, and sector information
• Risk Assessment: Computes volatility, Sharpe ratio, maximum drawdown, and current drawdown
• AI-Powered Recommendations: Uses Google Gemini AI to synthesize analysis and generate BUY/SELL/HOLD recommendations with detailed reasoning

Prerequisites

• Python 3.8 or higher
• Google Gemini API key (for AI synthesis)

Installation

1. Clone the repository:

cd financial_intellligence_system

2. Install dependencies:

pip install -r requirements.txt

3. Create a .env file in the project root directory:

# .env
GOOGLE_API_KEY=your_google_gemini_api_key_here

Note: You can obtain a Google Gemini API key from Google AI Studio.

How to Run

Command-Line Interface (CLI)

Run the analysis for any NYSE ticker symbol:

python test.py

When prompted, enter a ticker symbol (e.g., AMZN, GOOG, TSLA, AAPL, ORCL, MSFT, NVDA).

The system will:

1. Fetch historical data for the ticker
2. Perform technical and fundamental analysis
3. Calculate risk metrics
4. Generate an AI-powered recommendation with reasoning (using Gemini if available, otherwise offline analytics)

Note: If you enter an invalid ticker, the system will display a friendly error message instead of crashing.

API/UI Server

Start the FastAPI server with UI:

python -m uvicorn financial_intellligence_system.ui_server:app --reload --port 8000

The server will be available at:

• Web UI: http://127.0.0.1:8000
• API Endpoint: http://127.0.0.1:8000/api/analyze
• API Documentation:
  • Swagger UI: http://127.0.0.1:8000/api/docs
  • ReDoc: http://127.0.0.1:8000/api/redoc

Important: Make sure you have a .env file in the project root with:

GOOGLE_API_KEY=your_google_gemini_api_key_here

The system will work without the API key (using offline analytics), but Gemini AI recommendations require a valid key.

Troubleshooting: Import Issues

If you encounter import errors or "module not found" issues, it may be due to stale Python bytecode. Clean the cache:

python tools/clean_cache.py

This removes all __pycache__ folders and .pyc files recursively from the project.

API Endpoint

POST /analyze

Analyzes a stock ticker and returns comprehensive financial analysis.

Request Body:

{
  "ticker": "AAPL"
}

Response (200 OK):

{
  "ticker": "AAPL",
  "success": true,
  "llm_used": true,
  "analysis_date": "2024-01-15 14:30:00",
  "data_range": {
    "start": "2019-01-15",
    "end": "2024-01-15"
  },
  "fundamental_metrics": {
    "market_cap": 3000000000000,
    "pe_ratio": 28.5,
    "forward_pe": 25.2,
    "sector": "Technology"
  },
  "technical_indicators": {
    "current_price": 185.50,
    "rsi_14": 65.3,
    "macd_line": 2.15,
    "signal_line": 1.89,
    "trend": "Bullish"
  },
  "risk_metrics": {
    "max_drawdown": -0.25,
    "current_drawdown": -0.02,
    "volatility": 0.18,
    "sharpe_ratio": 1.45
  },
  "recommendation": "BUY",
  "risk_level": "Medium",
  "reasoning": "Strong technical indicators with bullish MACD trend and reasonable valuation...",
  "key_drivers": [
    "Bullish MACD trend",
    "RSI of 65.3 indicating healthy momentum",
    "Current drawdown of -2% showing recent recovery",
    "Sharpe ratio of 1.45 indicating good risk-adjusted returns"
  ],
  "final_report": "Full detailed report text..."
}

Note: The llm_used field indicates whether Gemini AI was used (true) or if the system fell back to offline analytics (false). The system gracefully falls back to analytics-based recommendations if the Gemini API is unavailable.

Error Responses:

• 400 Bad Request: Invalid or empty ticker symbol

  {
    "detail": "Ticker symbol cannot be empty"
  }

• 404 Not Found: No data found for the ticker

  {
    "detail": "No data found for ticker 'INVALID'"
  }

• 500 Internal Server Error: Server-side error during analysis

  {
    "detail": "Error message"
  }

Example API Usage

Using curl:

curl -X POST "http://127.0.0.1:8000/api/analyze" \
     -H "Content-Type: application/json" \
     -d '{"ticker": "AAPL"}'

Using Python requests:

import requests

response = requests.post(
    "http://127.0.0.1:8000/api/analyze",
    json={"ticker": "AAPL"}
)
result = response.json()
print(result["recommendation"])

System Features

Robust Error Handling

• Invalid Tickers: The system gracefully handles invalid ticker symbols and returns clear error messages instead of crashing.
• Missing Data: If yfinance returns no data, the system sets success=false and provides an error message.
• Gemini API Failures: If the Gemini API is unavailable (404, auth error, network error), the system automatically falls back to offline analytics-based recommendations using the same decision framework.

LLM Fallback

The system uses Google Gemini AI (gemini-2.0-flash with fallback to gemini-1.5-pro) when available, but works completely offline using computed analytics if:

• GOOGLE_API_KEY is not set in .env
• The Gemini API returns errors
• Network connectivity issues occur

The llm_used field in the response indicates whether AI was used or offline analytics were applied.