app.py

### FINAL APP.PY FOR HUGGING FACE USING THE IBM GRANITE MODEL ###

import streamlit as st
import torch
import fitz  # PyMuPDF
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.llms.huggingface_pipeline import HuggingFacePipeline
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
from langchain.prompts import PromptTemplate

# For Fairness Audit
import pandas as pd
from aif360.datasets import StandardDataset
from aif360.metrics import BinaryLabelDatasetMetric

# --- Page Configuration ---
st.set_page_config(
    page_title="Sahay AI 🇮🇳",
    page_icon="🤖",
    layout="wide",
    initial_sidebar_state="expanded"
)

# --- Caching for Performance ---
@st.cache_resource
def load_llm():
    """Loads the IBM Granite LLM, ensuring it runs on a GPU."""
    llm_model_name = "ibm-granite/granite-3.3-8b-instruct"
    
    # This check is crucial. The app will stop if no GPU is found.
    if not torch.cuda.is_available():
        raise RuntimeError("Hardware Error: This application requires a GPU to run the IBM Granite model. Please select a GPU hardware tier in your Space settings (e.g., T4 small).")

    model = AutoModelForCausalLM.from_pretrained(
        llm_model_name,
        torch_dtype=torch.bfloat16,
        load_in_4bit=True  # 4-bit quantization to save memory
    )
    tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
    
    pipe = pipeline(
        "text-generation", 
        model=model, 
        tokenizer=tokenizer, 
        max_new_tokens=512,
        temperature=0.1,
        device=0  # Force the pipeline to use the first available GPU
    )
    return HuggingFacePipeline(pipeline=pipe)

@st.cache_resource
def load_and_process_pdf(pdf_path):
    """Loads and embeds the PDF using IBM's multilingual model."""
    try:
        doc = fitz.open(pdf_path)
        text = "".join(page.get_text() for page in doc)
    except Exception as e:
        st.error(f"Error reading PDF: {e}. Ensure 'PMKisanSamanNidhi.PDF' is in the main project directory.")
        return None

    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=150)
    docs = text_splitter.create_documents([text])
    
    embedding_model = HuggingFaceEmbeddings(model_name="ibm-granite/granite-embedding-278m-multilingual")
    vector_db = FAISS.from_documents(docs, embedding_model)
    return vector_db

# --- Conversational Chain ---
def create_conversational_chain(_llm, _vector_db):
    prompt_template = """You are a polite AI assistant for the PM-KISAN scheme... (rest of prompt)"""
    QA_PROMPT = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
    memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True, output_key='answer')
    chain = ConversationalRetrievalChain.from_llm(
        llm=_llm, retriever=_vector_db.as_retriever(), memory=memory,
        return_source_documents=True, combine_docs_chain_kwargs={"prompt": QA_PROMPT}
    )
    return chain

# --- IBM AIF360 Fairness Audit ---
def run_fairness_audit():
    st.subheader("🤖 IBM AIF360 - Fairness Audit")
    df_display = pd.DataFrame({'gender_text': ['male', 'male', 'female', 'female']})
    df_for_aif = pd.DataFrame()
    df_for_aif['gender'] = df_display['gender_text'].map({'male': 1, 'female': 0})
    df_for_aif['favorable_outcome'] = [1, 1, 1, 1]
    aif_dataset = StandardDataset(df_for_aif, label_name='favorable_outcome', favorable_classes=[1],
                                  protected_attribute_names=['gender'], privileged_classes=[[1]])
    metric = BinaryLabelDatasetMetric(aif_dataset, unprivileged_groups=[{'gender': 0}], privileged_groups=[{'gender': 1}])
    spd = metric.statistical_parity_difference()
    st.metric(label="**Statistical Parity Difference (SPD)**", value=f"{spd:.4f}")

# --- Main Application UI ---
if __name__ == "__main__":
    
    with st.sidebar:
        st.image("https://upload.wikimedia.org/wikipedia/commons/5/51/IBM_logo.svg", width=100)
        st.title("🇮🇳 Sahay AI")
        st.markdown("An AI assistant for the **PM-KISAN** scheme, built on **IBM's Granite** foundation models.")
        if st.button("Run Fairness Audit", use_container_width=True):
            st.session_state.run_audit = True

    st.header("Chat with Sahay AI 💬")

    if st.session_state.get('run_audit', False):
        run_fair_audit()
        st.session_state.run_audit = False
    
    if "messages" not in st.session_state:
        st.session_state.messages = [{"role": "assistant", "content": "Welcome! How can I help you today?"}]

    if "qa_chain" not in st.session_state:
        with st.spinner("🚀 Waking up the IBM Granite Model... This may take several minutes on a GPU."):
            try:
                llm = load_llm()
                vector_db = load_and_process_pdf("PMKisanSamanNidhi.PDF")
                st.session_state.qa_chain = create_conversational_chain(llm, vector_db)
            except RuntimeError as e:
                st.error(e) # This will display the "Hardware Error" message from load_llm()
                st.stop()
    
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])

    if prompt := st.chat_input("Ask a question..."):
        st.session_state.messages.append({"role": "user", "content": prompt})
        with st.chat_message("user"):
            st.markdown(prompt)
        with st.chat_message("assistant"):
            with st.spinner("🧠 Thinking..."):
                result = st.session_state.qa_chain.invoke({"question": prompt})
                response = result["answer"]
                st.markdown(response)
        st.session_state.messages.append({"role": "assistant", "content": response})