app.py

# ✅ Import Libraries
import os
import gradio as gr
import datasets
from tqdm import tqdm
from transformers import AutoTokenizer
from langchain.docstore.document import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS  # ✅ FIXED IMPORT
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores.utils import DistanceStrategy
from smolagents import Tool, ToolCallingAgent, HfApiModel, DuckDuckGoSearchTool
from langchain_community.document_loaders import PyPDFLoader  # ✅ FIXED IMPORT

# ✅ REMOVE notebook_login()
# notebook_login()  # ❌ DELETE THIS LINE

# ✅ Step 2: Load PDF Documents for RAG
def load_documents(pdf_folder):
    """Loads PDFs and extracts text for RAG."""
    docs = []
    if not os.path.exists(pdf_folder):
        raise ValueError(f"❌ Error: The folder {pdf_folder} does not exist!")

    for file in os.listdir(pdf_folder):
        if file.endswith(".pdf"):
            file_path = os.path.join(pdf_folder, file)
            print(f"📂 Loading: {file_path}")
            loader = PyPDFLoader(file_path)  # ✅ FIXED
            docs.extend(loader.load())

    if not docs:
        raise ValueError("❌ Error: No valid PDFs found in the directory!")
    return docs

# ✅ Ensure PDF Folder Exists
pdf_folder = "/content"  # Change if needed
if os.path.exists(pdf_folder):
    documents = load_documents(pdf_folder)
else:
    documents = []

# ✅ Process Documents for Vector Search
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=200, chunk_overlap=20, add_start_index=True, strip_whitespace=True
)
docs_processed = text_splitter.split_documents(documents)

# ✅ Create FAISS Vector Database
embedding_model = HuggingFaceEmbeddings(model_name="thenlper/gte-small")
vector_db = FAISS.from_documents(
    documents=docs_processed,
    embedding=embedding_model,
    distance_strategy=DistanceStrategy.COSINE,
)

print("✅ FAISS Vector Database Successfully Created!")