app.py

import os
import feedparser
from flask import Flask, render_template
from huggingface_hub import HfApi, Repository
from langchain_huggingface import HuggingFaceInferenceClient
from langchain.vectorstores import Chroma
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.docstore.document import Document
import requests
import shutil

# Flask app setup
app = Flask(__name__)

# Hugging Face setup
HF_API_TOKEN = os.getenv("HF_API_TOKEN", "YOUR_HF_API_TOKEN")
HF_MODEL = "Qwen/Qwen-72B-Instruct"  # Qwen-72B model
REPO_ID = "your-username/news-rag-db"  # Replace with your HF repo ID
LOCAL_DB_DIR = "chroma_db"
client = HuggingFaceInferenceClient(model=HF_MODEL, api_key=HF_API_TOKEN)

# RSS feeds to fetch (example list)
RSS_FEEDS = [
    "http://rss.cnn.com/rss/cnn_topstories.rss",
    "https://feeds.bbci.co.uk/news/rss.xml",
    "https://www.npr.org/rss/rss.php?id=1001",
]

# Embedding model for vectorization
embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Initialize Chroma DB
vector_db = Chroma(persist_directory=LOCAL_DB_DIR, embedding_function=embedding_model)

# HfApi for Hugging Face Hub
hf_api = HfApi()

def fetch_rss_feeds():
    """Fetch news articles from RSS feeds."""
    articles = []
    for feed_url in RSS_FEEDS:
        feed = feedparser.parse(feed_url)
        for entry in feed.entries[:5]:  # Limit to 5 articles per feed for demo
            articles.append({
                "title": entry.get("title", "No Title"),
                "link": entry.get("link", ""),
                "description": entry.get("summary", entry.get("description", "No Description")),
                "published": entry.get("published", "Unknown Date"),
            })
    return articles

def summarize_article(text):
    """Summarize text using Qwen-72B via InferenceClient."""
    prompt = f"Summarize the following text in a concise manner:\n\n{text}"
    response = client.generate(prompt, max_new_tokens=100, temperature=0.7)
    return response.generated_text.strip()

def categorize_article(text):
    """Categorize text into positive, negative, or neutral using Qwen-72B."""
    prompt = f"Classify the sentiment of the following text as positive, negative, or neutral:\n\n{text}"
    response = client.generate(prompt, max_new_tokens=10, temperature=0.7)
    return response.generated_text.strip()

def process_and_store_articles(articles):
    """Process articles: summarize, categorize, vectorize, and store in RAG DB."""
    documents = []
    for article in articles:
        # Summarize and categorize
        summary = summarize_article(article["description"])
        category = categorize_article(article["description"])
        
        # Create document with metadata
        doc = Document(
            page_content=summary,
            metadata={
                "title": article["title"],
                "link": article["link"],
                "original_description": article["description"],
                "published": article["published"],
                "category": category,
            }
        )
        documents.append(doc)
    
    # Vectorize and store in Chroma DB
    vector_db.add_documents(documents)
    vector_db.persist()
    
    # Upload to Hugging Face Hub
    upload_to_hf_hub()

def upload_to_hf_hub():
    """Upload the Chroma DB to Hugging Face Hub."""
    if os.path.exists(LOCAL_DB_DIR):
        # Check if repo exists, create if not
        try:
            hf_api.create_repo(repo_id=REPO_ID, repo_type="dataset", exist_ok=True)
        except Exception as e:
            print(f"Error creating repo: {e}")
        
        # Upload all files in the DB directory
        for root, _, files in os.walk(LOCAL_DB_DIR):
            for file in files:
                local_path = os.path.join(root, file)
                remote_path = os.path.relpath(local_path, LOCAL_DB_DIR)
                hf_api.upload_file(
                    path_or_fileobj=local_path,
                    path_in_repo=remote_path,
                    repo_id=REPO_ID,
                    repo_type="dataset",
                    token=HF_API_TOKEN
                )
        print(f"Database uploaded to Hugging Face Hub: {REPO_ID}")

@app.route('/')
def index():
    """Render the Flask frontend with news articles."""
    articles = fetch_rss_feeds()
    process_and_store_articles(articles)
    
    # Retrieve summaries from the vector DB for display
    stored_docs = vector_db.similarity_search("news", k=len(articles))
    enriched_articles = []
    for doc in stored_docs:
        enriched_articles.append({
            "title": doc.metadata["title"],
            "link": doc.metadata["link"],
            "summary": doc.page_content,
            "category": doc.metadata["category"],
            "published": doc.metadata["published"],
        })
    
    return render_template("index.html", articles=enriched_articles)

# HTML template as a string (for simplicity)
HTML_TEMPLATE = """
<!DOCTYPE html>
<html>
<head>
    <title>News Feed</title>
    <style>
        body { font-family: Arial, sans-serif; margin: 20px; }
        .article { border-bottom: 1px solid #ccc; padding: 10px; }
        .title { font-size: 1.2em; }
        .summary { color: #555; }
        .category { font-style: italic; }
    </style>
</head>
<body>
    <h1>Latest News Feed</h1>
    {% for article in articles %}
    <div class="article">
        <div class="title"><a href="{{ article.link }}" target="_blank">{{ article.title }}</a></div>
        <div class="summary">{{ article.summary }}</div>
        <div class="category">Category: {{ article.category }}</div>
        <div>Published: {{ article.published }}</div>
    </div>
    {% endfor %}
</body>
</html>
"""

if __name__ == "__main__":
    # Save the HTML template to the templates folder
    os.makedirs("templates", exist_ok=True)
    with open("templates/index.html", "w") as f:
        f.write(HTML_TEMPLATE)
    
    # Clear existing DB for fresh start (optional)
    if os.path.exists(LOCAL_DB_DIR):
        shutil.rmtree(LOCAL_DB_DIR)
    
    # Run Flask app
    app.run(debug=True, host="0.0.0.0", port=5000)