app.py

import os
import requests
from groq import Groq
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.text_splitter import RecursiveCharacterTextSplitter
from PyPDF2 import PdfReader
from tempfile import NamedTemporaryFile

# Initialize Groq client
client = Groq(api_key=os.environ['GROQ_API_KEY'])

# Function to extract text from a PDF
def extract_text_from_pdf(pdf_file_path):
    pdf_reader = PdfReader(pdf_file_path)
    text = ""
    for page in pdf_reader.pages:
        page_text = page.extract_text()
        if page_text:
            text += page_text
    return text

# Function to split text into chunks
def chunk_text(text, chunk_size=500, chunk_overlap=50):
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=chunk_size, chunk_overlap=chunk_overlap
    )
    return text_splitter.split_text(text)

# Function to create embeddings and store them in FAISS
def create_embeddings_and_store(chunks, vector_db=None):
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    if vector_db is None:
        vector_db = FAISS.from_texts(chunks, embedding=embeddings)
    else:
        vector_db.add_texts(chunks)
    return vector_db

# Function to query the vector database and interact with Groq
def query_vector_db(query, vector_db):
    docs = vector_db.similarity_search(query, k=3)
    context = "\n".join([doc.page_content for doc in docs])
    chat_completion = client.chat.completions.create(
        messages=[
            {"role": "system", "content": f"Use the following context:\n{context}"},
            {"role": "user", "content": query},
        ],
        model="llama3-8b-8192",
    )
    return chat_completion.choices[0].message.content

# Function to convert Google Drive view link to downloadable link
def get_direct_download_link(view_url):
    if "drive.google.com/file/d/" in view_url:
        file_id = view_url.split("/file/d/")[1].split("/")[0]
        return f"https://drive.google.com/uc?export=download&id={file_id}"
    return None

# Function to download and save a PDF from a URL
def download_pdf_from_url(url):
    direct_url = get_direct_download_link(url)
    if not direct_url:
        return None
    response = requests.get(direct_url)
    if response.status_code == 200:
        temp_file = NamedTemporaryFile(delete=False, suffix=".pdf")
        temp_file.write(response.content)
        temp_file.close()
        return temp_file.name
    else:
        return None

# Function to process all documents and build vector DB
def process_documents(doc_links):
    vector_db = None
    for idx, link in enumerate(doc_links):
        print(f"📄 Processing document {idx + 1}...")
        pdf_path = download_pdf_from_url(link)
        if pdf_path:
            text = extract_text_from_pdf(pdf_path)
            chunks = chunk_text(text)
            vector_db = create_embeddings_and_store(chunks, vector_db=vector_db)
            print(f"✅ Document {idx + 1} processed.")
        else:
            print(f"❌ Failed to process document {idx + 1}")
    return vector_db

# Main callable function for Graido
def run_query_pipeline(doc_links, user_query):
    """
    Process documents and run a query. Returns LLM response.
    
    Args:
        doc_links (List[str]): List of Google Drive view links
        user_query (str): User's natural language query

    Returns:
        str: LLM-generated response based on document context
    """
    vector_db = process_documents(doc_links)
    
    if not vector_db:
        return "⚠️ No documents could be processed."
    
    if not user_query:
        return "⚠️ No user query provided."
    
    response = query_vector_db(user_query, vector_db)
    return response