app.py

from transformers import pipeline
from langchain.chains.question_answering import load_qa_chain
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain_community.document_loaders import PyPDFLoader, TextLoader
import gradio as gr

# Load local HF model pipeline
qa_pipeline = pipeline("text2text-generation", model="google/flan-t5-base")

# Load documents
def load_docs(file):
    if file.name.endswith(".pdf"):
        loader = PyPDFLoader(file.name)
    else:
        loader = TextLoader(file.name)
    return loader.load()

# Split text into chunks
def split_docs(documents):
    splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    return splitter.split_documents(documents)

# Embed documents and store in FAISS
def embed_docs(splits):
    embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
    return FAISS.from_documents(splits, embeddings)

# QA chain using local pipeline
def qa_chain(vectorstore, query):
    from langchain.llms import HuggingFacePipeline
    llm = HuggingFacePipeline(pipeline=qa_pipeline)
    chain = load_qa_chain(llm, chain_type="stuff")
    docs = vectorstore.similarity_search(query)
    return chain.run(input_documents=docs, question=query)

# Gradio interface
def analyze(file, question):
    docs = load_docs(file)
    splits = split_docs(docs)
    vectorstore = embed_docs(splits)
    answer = qa_chain(vectorstore, question)
    return answer

interface = gr.Interface(
    fn=analyze,
    inputs=[
        gr.File(label="Upload Document"),
        gr.Textbox(label="Enter Your Question")
    ],
    outputs="text",
    title="📄 RAG-Powered Document Analyzer",
    description="Upload a document and ask questions about its content!"
)

if __name__ == "__main__":
    interface.launch(share=True)