app.py

import os
import requests
import gradio as gr
from cerebras.cloud.sdk import Cerebras
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.schema import Document
import numpy as np
from langchain_community.document_loaders import TextLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from sentence_transformers import SentenceTransformer

# Initialize Cerebras API client

Facts = os.getenv("Facto")
client = Cerebras(api_key= Facts)

Newskey = os.getenv("News")
# Function to fetch latest news articles from NewsAPI
def get_latest_news(query):
    api_key = Newskey
    url = f"https://newsapi.org/v2/everything?q={query}&apiKey={api_key}"
    response = requests.get(url)
    data = response.json()
    return [(article["title"], article["url"], article["source"]["name"]) for article in data.get("articles", [])[:5]]

# Function to update fact_checks.txt with new user input (overwrites previous content)
def update_fact_checks_file(query):
    with open("fact_checks.txt", "w", encoding="utf-8") as file:
        file.write(f"{query}\n")

# Function to create a FAISS retriever dynamically
def create_faiss_retriever():
    if not os.path.exists("fact_checks.txt"):
        open("fact_checks.txt", "w").close()  # Create file if it doesn't exist
    
    loader = TextLoader("fact_checks.txt")
    documents = loader.load()
    
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=400, chunk_overlap=50)
    docs = text_splitter.split_documents(documents)
    
    embedding_model = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    vector_store = FAISS.from_documents(docs, embedding_model)
    
    return vector_store.as_retriever(search_type="similarity", search_kwargs={"k": 8})

# Function to clear the fact_checks.txt file after execution
def clear_fact_checks_file():
    open("fact_checks.txt", "w").close()

# Function to perform fact-checking with Llama 3.3
def fact_check_with_llama3(query):
    # Save query to fact_checks.txt
    update_fact_checks_file(query)

    # Reload FAISS index with new data
    retriever = create_faiss_retriever()

    # Retrieve relevant facts from FAISS
    retrieved_docs = retriever.invoke(query)
    retrieved_texts = [doc.page_content for doc in retrieved_docs]

    # Fetch real-time news
    news = get_latest_news(query)

    # Combine all retrieved context
    context_text = "\n".join(retrieved_texts)
    
    # Construct prompt for Llama 3.3
    prompt = f"""
    Claim: {query}
    Context: {context_text}
    Based on the provided context, determine whether the claim is True, False, or Misleading. Provide a concise explanation and cite relevant sources. Don't mention any instance of your knowledge cut-off.
    """
    
    # Call Llama 3.3 API
    stream = client.chat.completions.create(
        messages=[{"role": "system", "content": prompt}],
        model="llama-3.3-70b",
        stream=True,
        max_completion_tokens=512,
        temperature=0.2,
        top_p=1
    )
    
    # Generate AI response
    result = "".join(chunk.choices[0].delta.content or "" for chunk in stream)
    
    # Format results with sources
    sources = "\n".join([f"{title} ({source}): {url}" for title, url, source in news])

    # Clear the file after execution
    clear_fact_checks_file()

    return result, sources if sources else "No relevant sources found."

# Gradio Interface
def fact_check_interface(query):
    response, sources = fact_check_with_llama3(query)
    return response, sources

gui = gr.Interface(
    fn=fact_check_interface,
    inputs=gr.Textbox(placeholder="Enter a claim to fact-check"),
    outputs=[gr.Textbox(label="Fact-Check Result"), gr.Textbox(label="Sources")],
    title="Facto - AI Fact-Checking System",
    description="Enter a claim, and the system will verify it using Llama 3.3 and external knowledge sources, citing relevant sources."
)

gui.launch(debug=True)