app.py

import streamlit as st
import os
import re

from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_core.prompts import PromptTemplate
from langchain_community.embeddings import HuggingFaceEmbeddings

from dotenv import load_dotenv
from unstructured.partition.auto import partition

# ==================== ENV SETUP ====================
load_dotenv()

if not os.getenv("GOOGLE_API_KEY"):
    st.error("❌ GOOGLE_API_KEY not found. Add it in Hugging Face Secrets.")
    st.stop()

# Disable inference for safety (you can remove this in Docker if you want full inference)
os.environ["UNSTRUCTURED_DISABLE_INFERENCE"] = "true"


# ==================== QUESTION SPLITTER ====================
def split_questions(text):
    text = text.replace("\n", " ").strip()

    # Split on ? OR . only if the dot is NOT part of a number like "1."
    questions = re.split(
        r'(?<!\d)(?<=[?.])\s*',
        text
    )

    return [q.strip() for q in questions if q.strip()]



# ==================== PROMPT ====================
PROMPT = PromptTemplate(
    template="""
Answer the question using ONLY the given context.
Respond in the SAME language as the question.
If the answer is not present, say:
"Answer is not available in the context."

Context:
{context}

Question:
{question}

Answer:
""",
    input_variables=["context", "question"]
)


# ==================== DOCUMENT INGESTION ====================
def extract_text_unstructured(uploaded_files):
    full_text = ""

    for file in uploaded_files:
        with open(file.name, "wb") as f:
            f.write(file.getbuffer())

        elements = partition(
            filename=file.name,
            strategy="fast"
        )

        file_text = "\n".join(el.text for el in elements if el.text)
        full_text += f"\n\n--- Source: {file.name} ---\n\n{file_text}"

        os.remove(file.name)

    return full_text


# ==================== CHUNKING ====================
def get_text_chunks(text):
    splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000,
        chunk_overlap=200
    )
    return splitter.split_text(text)


# ==================== EMBEDDINGS ====================
@st.cache_resource
def load_embeddings():
    return HuggingFaceEmbeddings(
        model_name="paraphrase-multilingual-MiniLM-L12-v2"
    )


# ==================== VECTOR STORE ====================
def get_vector_store(text_chunks):
    embeddings = load_embeddings()
    db = FAISS.from_texts(text_chunks, embedding=embeddings)
    db.save_local("faiss_index")


# ==================== GEMINI ====================
def ask_gemini(context, question):
    llm = ChatGoogleGenerativeAI(
        model="gemini-2.5-flash",
        temperature=0.3
    )

    response = llm.invoke(
        PROMPT.format(context=context, question=question)
    )
    return response.content


# ==================== USER QUERY ====================
def user_input(user_question):
    if not os.path.exists("faiss_index"):
        st.warning("Please upload and process files first.")
        return

    embeddings = load_embeddings()
    db = FAISS.load_local(
        "faiss_index",
        embeddings,
        allow_dangerous_deserialization=True
    )

    questions = split_questions(user_question)

    for idx, question in enumerate(questions, start=1):
        st.markdown(f"###  Question {idx}")
        st.write(question)

        docs = db.similarity_search(question, k=3)

        if not docs:
            st.write("Answer is not available in the context.")
            st.divider()
            continue

        context = "\n\n".join(doc.page_content for doc in docs)

        with st.spinner("Thinking..."):
            answer = ask_gemini(context, question)

        st.markdown("**✅ Reply:**")
        st.write(answer)
        st.divider()


# ==================== CACHE ====================
def clear_cache():
    st.cache_resource.clear()
    st.cache_data.clear()


# ==================== STREAMLIT UI ====================
def main():
    st.set_page_config(page_title="Chat PDF")
    st.header("📘 Syllabus RAG System")

    user_question = st.text_input("Ask a question from the uploaded documents")

    if user_question:
        user_input(user_question)

    with st.sidebar:
        st.title("Menu")

        pdf_docs = st.file_uploader(
            "Upload files",
            type=["pdf", "txt", "md", "docx", "html"],
            accept_multiple_files=True
        )

        if st.button("Submit & Process"):
            if not pdf_docs:
                st.warning("Please upload at least one file.")
                return

            with st.spinner("Processing files..."):
                raw_text = extract_text_unstructured(pdf_docs)
                chunks = get_text_chunks(raw_text)
                get_vector_store(chunks)
                st.success("✅ Files processed successfully!")

        if st.button("Clear Cache"):
            clear_cache()
            st.success("Cache cleared successfully!")


if __name__ == "__main__":
    main()