📈 AI Financial Analyst Agent

An agentic AI workflow designed to perform automated stock research, sentiment analysis, and paper trading with a built-in Human-in-the-Loop approval system.

🚀 Overview

This agent leverages LangGraph to coordinate a multi-step financial analysis process. It fetches live news and market data, uses Google Gemini to synthesize a sentiment-based recommendation, and allows you to approve or reject trades directly from a Streamlit dashboard.

---

🛠️ Technology Stack

• Workflow Engine: LangGraph (for stateful, cyclic graphs)
• AI Brain: Google Gemini 1.5 Flash (Sentiment analysis & decision making)
• Search Tool: Tavily AI (Real-time financial news search)
• Market Data: yfinance (Live stock prices and P/E ratios)
• Trading API: Alpaca Markets (Paper trading execution)
• Frontend: Streamlit (Interactive dashboard)

---

🔄 Agentic Workflow (The Process)

The agent moves through a defined set of "Nodes" in a specific order:

1. Research Node

• News Search: Calls Tavily to find the top 5 most relevant news headlines for the target ticker (e.g., "NVDA stock performance today").
• Price Fetch: Uses yfinance to retrieve the current market price and the trailing P/E ratio.
• Timestamping: Marks the exact time data was fetched to ensure fresh analysis.

2. Analysis Node

• Data Synthesis: Consolidates news headlines and price metrics into a prompt.
• LLM Reasoning: Gemini analyzes the news context against the price data.
• Recommendation: Gemini outputs a JSON response containing a summary, sentiment (POSITIVE/NEGATIVE/NEUTRAL), and a suggested action (BUY/SELL/HOLD).

3. Human-in-the-Loop (Interrupt)

• Stale Data Check: If more than 10 minutes (configurable) have passed since the data was fetched, the agent forces a loop back to the Research Node to prevent trading on old news.
• Approval Window: The workflow pauses. The user must review the AI's summary and recommendation.
• Decision: The user clicks "Approve" to proceed to execution or "Reject" to end the run.

4. Action Node

• Execution: If approved, the agent calls the Alpaca API to place a Market Order in your paper trading account.
• Validation: Captures the Order ID for tracking.

5. Complete

• Finalizes the state and displays a summary of the entire run including news, price, sentiment, and trade status.

---

⚙️ Setup & Execution

1. Requirements

Install dependencies using pip:

pip install langgraph langchain-google-genai tavily-python yfinance alpaca-trade-api python-dotenv streamlit

2. Environment Variables

Create a .env file with your API keys:

TAVILY_API_KEY=your_key
GOOGLE_API_KEY=your_key
ALPACA_API_KEY=your_key
ALPACA_SECRET_KEY=your_key
ALPACA_BASE_URL=https://paper-api.alpaca.markets

3. Running the App

Launch the interactive dashboard:

streamlit run app.py

---

🛡️ Safety Features

• Human-in-the-Loop: No trades are ever placed without explicit manual approval in the UI.
• Stale Data Protection: Automatically detects if market conditions might have changed since the start of the analysis and triggers a re-fetch.
• Paper Trading: Defaulted to Alpaca's paper API to ensure no real money is at risk during testing.