app.py

import os
import streamlit as st
from groq import Groq
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import FAISS
from langchain.text_splitter import CharacterTextSplitter
from langchain.document_loaders import TextLoader, PyPDFLoader
from langchain.chains import RetrievalQA
from langchain.llms.base import LLM
from tempfile import NamedTemporaryFile

# Load Groq API key from environment variable
GROQ_API_KEY = os.getenv("GROQ_API_KEY")

# Initialize Groq client
groq_client = Groq(api_key=GROQ_API_KEY)

# Define a basic LLM wrapper for Groq
class GroqLLM(LLM):
    def __init__(self, model_name="llama3-8b-8192"):
        self.model_name = model_name

    def _call(self, prompt, stop=None):
        response = groq_client.chat.completions.create(
            model=self.model_name,
            messages=[{"role": "user", "content": prompt}],
        )
        return response.choices[0].message.content.strip()

    @property
    def _llm_type(self):
        return "groq_llm"

# Streamlit UI
st.title("📚 RAG App with Groq + HuggingFace + Streamlit")
st.write("Upload a PDF or TXT file, ask a question, and get answers powered by RAG.")

uploaded_file = st.file_uploader("Upload your document", type=["pdf", "txt"])

if uploaded_file:
    with NamedTemporaryFile(delete=False) as tmp_file:
        tmp_file.write(uploaded_file.read())
        tmp_path = tmp_file.name

    # Load document
    if uploaded_file.type == "application/pdf":
        loader = PyPDFLoader(tmp_path)
    else:
        loader = TextLoader(tmp_path)

    docs = loader.load()

    # Split into chunks
    splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
    texts = splitter.split_documents(docs)

    # Create embeddings and FAISS index
    embeddings = HuggingFaceEmbeddings()
    db = FAISS.from_documents(texts, embeddings)

    # RAG chain
    retriever = db.as_retriever()
    qa_chain = RetrievalQA.from_chain_type(llm=GroqLLM(), retriever=retriever)

    # Input box
    query = st.text_input("Ask something about the document:")

    if query:
        result = qa_chain.run(query)
        st.markdown("### 🧠 Answer:")
        st.success(result)