FinAgent Pro: AI Financial Analysis Agent

FinAgent Pro is a sophisticated AI-powered agent designed to perform financial analysis of U.S. companies. By leveraging a multi-step, tool-based workflow orchestrated by LangGraph, this agent can retrieve SEC filings, calculate technical stock indicators, and synthesize the information into a concise report. The interactive user interface is built with Streamlit.

---

Features

• Automated SEC Filing Retrieval: Automatically finds a company's CIK and fetches the latest annual (10-K) or quarterly (10-Q) reports directly from the SEC EDGAR database.
• Technical Indicator Calculation: Gathers historical stock data using yfinance and calculates key technical indicators like RSI, MACD, and SMA using pandas-ta.
• Web Content Parsing: Intelligently reads the content of SEC filings from the web for analysis.
• Stateful, Multi-Step Workflow: Utilizes LangGraph to create a robust, sequential chain of actions, ensuring a reliable analysis process.
• Conversational Interface: A simple and intuitive chat interface powered by Streamlit allows users to make requests in natural language.
• Error Handling: The agent is designed to identify and report tool execution errors, providing clear feedback to the user.

---

How It Works: The Agent's Architecture

FinAgent Pro operates as a stateful agent graph. When a user provides a company name, the agent initiates a sequential workflow managed by LangGraph:

1. Get Company Info
   Calls get_company_info to find the company's CIK and ticker from SEC data.

2. Fetch Filings & Stock Data

   • Calls get_latest_sec_filings to retrieve the most recent 10-K filing (in .txt format).
   • Calls get_stock_data to fetch the last 365 days of price history.

3. Process Data

   • Sends stock data to calculate_technical_indicators.
   • Sends filing URL to read_webpage to extract text.

4. Synthesize Final Report
   The LLM generates a final summary based on the gathered data.

This process includes robust error handling to inform the user about any failed steps.

---

Technology Stack

• Frontend: Streamlit
• AI Orchestration: LangChain, LangGraph
• LLM Provider: OpenRouter (e.g., DeepSeek, GPT-4o, etc.)
• Data Sources:
  • SEC EDGAR API
  • Yahoo Finance (yfinance)
• Libraries: langchain-openai, pandas, pandas-ta, requests, beautifulsoup4

---

Setup and Installation

1. Prerequisites

• Python 3.11+
• pip and venv

2. Clone the Repository

git clone https://github.com/nam-htran/FinAgent-RAG
cd FinAgent-RAG

3. Set Up a Virtual Environment

# macOS/Linux
python3 -m venv venv
source venv/bin/activate

# Windows
python -m venv venv
.env\Scripts\activate

4. Install Dependencies

pip install -r requirements.txt

5. Configure Environment Variables

Copy the example file and update your credentials:

cp .env.example .env

Edit .env:

# .env

# Get your key from https://openrouter.ai/keys
OPENROUTER_API_KEY="sk-or-..."

# Required for SEC API access
SEC_USER_AGENT="YourFullName YourEmail@example.com"

The SEC requires a valid User-Agent. Generic agents like python-requests will be blocked.

6. Run the Application

streamlit run app.py

Your browser should now open the app interface.

---

Usage

Once the app is running, type a query in the chat such as:

• Analyze the latest annual report for Microsoft
• Give me a financial and technical summary for Tesla
• Apple 2024 financial report analysis

The agent will retrieve data, analyze it, and return a summarized report.

---

Roadmap & Future Improvements

• Data Visualization: Add charts with matplotlib or plotly.
• Streaming Responses: Show real-time steps of agent reasoning.
• News Analysis Tool: Include company news with sentiment scoring.
• Advanced Caching: Speed up repeated queries with caching.
• Model Comparison UI: Let users test and compare LLM outputs.

---