app.py

# =========================
# IMPORTS
# =========================
import os
import tempfile
import gradio as gr

from groq import Groq
from duckduckgo_search import DDGS

from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import Chroma


# =========================
# CONFIG
# =========================
GROQ_API_KEY = os.getenv("GROQ_API_KEY")  # HF SECRET KEY

client = Groq(api_key=GROQ_API_KEY)

vectorstore = None
retriever = None


# =========================
# PROMPT
# =========================
def build_prompt(context, question):
    return f"""
You are an expert AI assistant.

Use ONLY the context below.
If answer is not present, say "Not found in document".

CONTEXT:
{context}

QUESTION:
{question}

ANSWER:
"""


# =========================
# WEB SEARCH (FALLBACK)
# =========================
def web_search(query):
    results = []
    with DDGS() as ddgs:
        for r in ddgs.text(query, max_results=3):
            results.append(r["body"])
    return "\n\n".join(results)


# =========================
# PROCESS PDF (HF SAFE)
# =========================
def process_pdf(file):

    global vectorstore, retriever

    # safe file handling
    file_path = file.name

    loader = PyPDFLoader(file_path)
    documents = loader.load()

    splitter = RecursiveCharacterTextSplitter(
        chunk_size=600,
        chunk_overlap=100
    )

    chunks = splitter.split_documents(documents)

    embedding_model = HuggingFaceEmbeddings(
        model_name="sentence-transformers/all-MiniLM-L6-v2"
    )

    vectorstore = Chroma.from_documents(
        documents=chunks,
        embedding=embedding_model
    )

    retriever = vectorstore.as_retriever(search_kwargs={"k": 4})

    return "✅ PDF successfully processed"


# =========================
# RAG ENGINE (HYBRID)
# =========================
def ask_rag(query):

    global retriever

    if retriever is None:
        return "⚠️ Please upload a PDF first."

    docs = retriever.invoke(query)
    pdf_context = "\n\n".join([d.page_content for d in docs])

    # hybrid fallback
    if len(pdf_context.strip()) < 50:
        web_context = web_search(query)
        context = pdf_context + "\n\nWEB:\n" + web_context
    else:
        context = pdf_context

    prompt = build_prompt(context, query)

    response = client.chat.completions.create(
        messages=[{"role": "user", "content": prompt}],
        model="llama3-8b-8192"
    )

    return response.choices[0].message.content


# =========================
# CHAT FUNCTION (FIXED FORMAT)
# =========================
def chat(user_message, history):

    response = ask_rag(user_message)

    if history is None:
        history = []

        history.append((user_message, response))

    return history, history


# =========================
# UI (HUGGING FACE SAFE)
# =========================
with gr.Blocks() as app:

    gr.Markdown("# 🧠 Hybrid RAG Chatbot (PDF + Web)")

    file = gr.File(label="Upload PDF")
    status = gr.Textbox(label="Status")

    chatbot = gr.Chatbot()  # IMPORTANT FIX
    msg = gr.Textbox(placeholder="Ask your question...")
    state = gr.State([])

    file.change(process_pdf, file, status)
    msg.submit(chat, [msg, state], [chatbot, state])

app.launch()