🧠 Introduction

In this post, we’ll build a real-time pipeline that combines FinBERT, OpenAI GPT, and a frontend in Astro to analyze news sentiment and predict stock momentum. This is a simplified version of what large trading desks use — and you can do it with just a few lines of code and free APIs.

⚙️ Architecture Overview

• Data Source: News headlines from Finnhub
• Sentiment Engine: FinBERT + OpenAI GPT
• Frontend: Astro static site + Markdown for blogs
• Optional: Store results in SQLite or Supabase for history

🛠️ Step 1: Get Stock News with Python

import requests
url = "https://finnhub.io/api/v1/news?category=general&token=YOUR_TOKEN"
news = requests.get(url).json()

🤖 Step 2: Run FinBERT for Sentiment

from transformers import pipeline

sentiment = pipeline("sentiment-analysis", model="ProsusAI/finbert")
for item in news:
    print(sentiment(item["headline"]))

🧠 Step 3: Boost with OpenAI GPT (Optional)

import openai

openai.api_key = "YOUR_API_KEY"
prompt = f"What is the sentiment of this stock news: '{{headline}}'?"

response = openai.ChatCompletion.create(
  model="gpt-4",
  messages=[{{"role": "user", "content": prompt}}]
)

🌐 Step 4: Publish Results with Astro

1. Create .md posts dynamically
2. Add live sentiment scores
3. Style with Tailwind for clean UI

📈 Bonus: Highlight Bullish/Bearish Stocks

You can tag each news item with:

• 📈 Bullish → “buy momentum”
• 📉 Bearish → “risk alert”

🎯 Conclusion

This pipeline forms the base of a larger real-time trading system. In future posts, we’ll connect this to live price prediction models, portfolio alerts, and agentic decision-making.

📎 Related Reading

• FinBERT: https://huggingface.co/ProsusAI/finbert
• Astro: https://astro.build/
• OpenAI API: https://platform.openai.com