app.py

import streamlit as st
import os
from llama_index.core import (
    VectorStoreIndex,
    SimpleDirectoryReader,
    Settings,
    PromptTemplate,
    QueryBundle,
)
from llama_index.llms.gemini import Gemini
from llama_index.embeddings.gemini import GeminiEmbedding
from llama_index.core import get_response_synthesizer
from llama_index.core.node_parser import SemanticSplitterNodeParser
from llama_index.core.retrievers import VectorIndexRetriever
from llama_index.core.query_engine import RetrieverQueryEngine
from llama_index.core.indices.query.query_transform import HyDEQueryTransform
from llama_index.core.postprocessor import SentenceTransformerRerank
from llama_index.core import load_index_from_storage
from llama_index.core import StorageContext
from llama_index.core.retrievers import QueryFusionRetriever
from dotenv import load_dotenv
import logging
import google.generativeai as genai
from pathlib import Path

load_dotenv()

# Set logging level
logging.basicConfig(level=logging.INFO)

# Configure Gemini Pro
genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))

model_gemini_pro_vision = "gemini-pro-vision"
model_gemini_pro = "gemini-pro"


# Configure Gemini models
Settings.llm = Gemini(models=model_gemini_pro, api_key=os.getenv("GOOGLE_API_KEY"))
Settings.embed_model = GeminiEmbedding(
    model_name="models/embedding-001",
    api_key=os.getenv("GOOGLE_API_KEY")
)


# Function to create a Semantic Splitter Node Parser
def create_semantic_splitter_node_parser():
    """Creates a semantic splitter."""
    return SemanticSplitterNodeParser(
        buffer_size=1, breakpoint_percentile_threshold=95, embed_model=Settings.embed_model
    )


def load_and_index_pdf(pdf_path):
    """Loads and index the pdf.
    
    Args : 
    pdf_path (str) : The path to the pdf file
    
    Returns : 
    index (llama_index.core.VectorStoreIndex): The vector index
    """
    try:
         logging.info(f"Loading PDF document from: {pdf_path}")
         documents = SimpleDirectoryReader(input_files=[pdf_path]).load_data()
         if documents:
            logging.info("Creating semantic splitter")
            node_parser = create_semantic_splitter_node_parser()
            nodes = node_parser.get_nodes_from_documents(documents)
            logging.info("Creating vector store index")
            index = VectorStoreIndex(nodes=nodes)
            return index
         else:
            logging.warning("No documents found in the PDF")
            return None
    except Exception as e:
        logging.error(f"Error loading and indexing PDF: {e}")
        return None


def create_rag_pipeline(index):
    """Creates a RAG pipeline for translation.
    
     Args : 
    index (llama_index.core.VectorStoreIndex): The vector index.
    
    Returns : 
    query_engine(llama_index.core.query_engine.RetrieverQueryEngine): The query engine
    """

    logging.info("Initializing RAG Pipeline components")
    # setup retriever

    retriever = VectorStoreIndex(
            index.nodes,
            ).as_retriever(similarity_top_k=5)


     # setup query transformer
    hyde_query_transform = HyDEQueryTransform(llm=Settings.llm)

      # setup reranker
    reranker = SentenceTransformerRerank(top_n=3, model="BAAI/bge-reranker-base")

    # response_synthesizer
    response_synthesizer = get_response_synthesizer(
            response_mode="refine",
        )
    
    # setup query engine
    query_engine = RetrieverQueryEngine(
        retriever=retriever,
        response_synthesizer=response_synthesizer,
        node_postprocessors=[reranker],
        query_transform= hyde_query_transform
    )

    logging.info("RAG Pipeline is configured.")
    return query_engine

def translate_text(french_text, query_engine):
    """Translates french text to Yipunu using a highly optimized RAG.
    
    Args : 
    french_text (str): The french text to translate.
    query_engine (llama_index.core.query_engine.RetrieverQueryEngine): The query engine.
    
    Returns:
     (str): The yipunu translation or an error message.
    """

    try:
        logging.info(f"Initiating translation of: {french_text}")

        template = (
            "Tu es un excellent traducteur du français vers le yipunu. Tu traduis le texte sans donner d'explication. "
            "Texte: {french_text} "
            "Traduction:"
        )

        prompt_template = PromptTemplate(template)
        query_bundle = QueryBundle(french_text, custom_prompt=prompt_template)
        response = query_engine.query(query_bundle)
        logging.info(f"Translation Result: {response.response}")
        return response.response
    except Exception as e:
        logging.error(f"Error during translation: {e}")
        return f"Error during translation: {str(e)}"
    


def main():
    """Main function for streamlit app."""
    
    st.title("French to Yipunu Translation App")
    
    # Construct the path to the PDF in the data folder
    default_pdf_path = Path("data/parlons_yipunu.pdf")
    
    # Check if the default pdf_file exists.
    if default_pdf_path.exists():
        index = load_and_index_pdf(str(default_pdf_path))
        if index:
            query_engine = create_rag_pipeline(index)
            french_text = st.text_area("Enter French Text:", "Ni vosi yipunu")
            if st.button("Translate"):
                translation = translate_text(french_text, query_engine)
                st.success(f"Yipunu Translation: {translation}")
    else:
            # PDF File Upload
        uploaded_file = st.file_uploader("Upload a PDF file containing the Punu grammar:", type="pdf")
        if uploaded_file is not None:
            # Save file to a temporary location
            temp_file_path = Path("temp_file.pdf")
            with open(temp_file_path, "wb") as f:
                f.write(uploaded_file.read())

            index = load_and_index_pdf(str(temp_file_path))
            if index:
                query_engine = create_rag_pipeline(index)
                french_text = st.text_area("Enter French Text:", "Ni vosi yipunu")
                if st.button("Translate"):
                    translation = translate_text(french_text, query_engine)
                    st.success(f"Yipunu Translation: {translation}")

            # Clean up temp files
            os.remove(temp_file_path)
        else:
            st.info("Please upload a pdf containing the punu grammar.")
    


if __name__ == "__main__":
    main()