LLM Demo News

Overview

LLM Demo News is a demonstration project showcasing the integration of financial news sentiment analysis using a Large Language Model (LLM). This project fetches market-related news articles, classifies their sentiment (positive, negative, neutral), and categorizes them by industry (e.g., Technology, Healthcare, etc.). It includes a backend powered by FastAPI and a React-based frontend for real-time visualization.

You will need access to an LLM, like open AI and a free finnhub API token, which you can get here: https://finnhub.io/

Features

• News Fetching: Retrieves market news from Finnhub.
• Sentiment Analysis: Uses an LLM to classify article sentiment from -5 (negative) to 5 (positive).
• Industry Categorization: Assigns articles to specific industries based on content.
• Real-Time Dashboard: Displays a timeline of sentiment scores and categorized articles in a browser-based UI.

Tech Stack

• Backend: FastAPI, Python
• Frontend: React, Chart.js
• Database: SQLite
• LLM Integration: Local endpoint for LLM (compatible with OpenAI APIs)
• APIs:
  • Finnhub API for news data
  • Custom LLM API for sentiment analysis

Setup

Prerequisites

Ensure the following tools are installed on your system:

• Python 3.11+
• Node.js 16+
• npm 8+
• pip (Python package manager)

Backend Installation

1. Clone the Repository

git clone https://github.com/castai/llm_demo_news
cd llm-demo-new

2. Create a Virtual Environment

python -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Configure secrets in config.yaml

llm:
  url: "http://localhost:8090/v1/"
  api_key: "your-llm-api-key"

finnhub:
  api_key: "your-finnhub-api-key"

5. Run the Server

./scripts/start_server.sh

The backend will be available at http://127.0.0.1:8000

Frontend Installation

1. Navigate to the dashboard

cd news-dashboard

2. Install Dependancies

npm install

3. Start the development server

npm run start

The frontend will be available at http://localhost:3000.

Usage

1. Start both the backend and frontend servers.
2. Open the frontend in your browser (http://localhost:3000).
3. Use the controls to:
   • Start/Stop polling news articles.
   • Start/Stop classifying articles.
   • Reset classifications.
4. View the sentiment timeline and detailed news articles on the dashboard.