app.py

import gradio as gr
from src.rag.question_answerer import QuestionAnsweringBot
from src.rag.retriever import Retriever


def ask_question(
    question: str,
    model: str,
    api_key: str,
    semantic_usage: float,
    initial_top_n: int,
    final_top_n: int,
) -> tuple[str, str]:
    """
    Handles question input from the user and returns the answer with relevant context chunks.

    Args:
        question (str): User's question.
        api_key (str): The API key for LiteLLM.
        semantic_usage (float): Weight for semantic usage in retrieval (0-1).

    Returns:
        tuple[str, str]: The answer and the relevant context chunks.
    """
    # Set the API key and model name
    qa_bot.model = model
    qa_bot.api_key = api_key

    # Get the answer and context chunks
    answer, contexts = qa_bot.answer_question(
        question=question,
        bm25_weight=(1 - semantic_usage),
        initial_top_n=initial_top_n,
        final_top_n=final_top_n,
    )

    # Format contexts as a single string
    formatted_contexts = "\n".join(
        [
            f"Context {i}:\n{chunk}" + "\n-----------------------------------\n"
            for i, chunk in enumerate(contexts, 1)
        ]
    )

    return answer, formatted_contexts


if __name__ == "__main__":
    # Initialize retriever and bot
    retriever = Retriever(chunked_dir="data/chunked")
    qa_bot = QuestionAnsweringBot(retriever=retriever)

    # Gradio UI
    with gr.Blocks() as demo:
        # Add custom CSS for Montserrat font, button styling, and purple slider/checkbox
        demo.css = """
            @import url('https://fonts.googleapis.com/css2?family=Montserrat:wght@400;600&display=swap');

            * {
                font-family: 'Montserrat', sans-serif !important;
            }

            #custom-button {
                color: #9d00ff !important;
                background-color: transparent !important;
                border: 2px solid #9d00ff !important;
                border-radius: 5px !important;
                padding: 10px 20px !important;
                font-weight: bold !important;
                font-size: 16px !important;
                cursor: pointer !important;
            }

            #custom-button:hover {
                background-color: #9d00ff !important;
                color: white !important;
            }


            /* Slider bar background */
            input[type="range"]::-webkit-slider-runnable-track{
                background: #e4e4e4;
                border-radius: 3px;
                height: 8px;
            }

            /* Circle */
            input[type="range"]::-webkit-slider-thumb {
                -webkit-appearance: none;
                appearance: none;
                width: 15px;
                height: 15px;
                background: #9d00ff;
                border-radius: 50%;
                cursor: pointer;
            }
        """

        # Short Description
        gr.Markdown(
            """
            <h1 style='text-align: center; color: #9d00ff;'>RAG Contextual Answering</h1>
            <p style='text-align: center;'>This tool allows you to ask questions and receive contextual answers 
            with relevant information from the files.</p>
            """
        )

        # Input Section
        with gr.Group():
            with gr.Row():
                model = gr.Textbox(
                    value="groq/llama3-8b-8192",
                    label="Model Name",
                    placeholder="Enter your model name here",
                )
                api_key = gr.Textbox(
                    label="API Key",
                    placeholder="Enter your API key here",
                    type="password",
                )

            question = gr.Textbox(
                value="Which survival instincts prey have?",
                label="Question",
                placeholder="Type your question here...",
            )
            semantic_usage = gr.Slider(
                label="Semantic usage (0 - only key phrases search, 1 - only semantic search)",
                minimum=0, maximum=1, value=0.5, step=0.001,
            )
            with gr.Row():
                initial_top_n = gr.Slider(
                    label="BM25 & Semantic Search Top N",
                    minimum=1, maximum=100, value=50, step=1,
                )
                final_top_n = gr.Slider(
                    label="Reranker Top N",
                    minimum=1, maximum=100, value=5, step=1,
                )

        # Submit Button
        submit_button = gr.Button("Get Answer", elem_id="custom-button")

        # Answer and Context Sections
        answer = gr.Textbox(label="Answer", interactive=False, lines=5)
        chunks_response = gr.Textbox(
            label="Context Chunks",
            interactive=False,
        )

        # Button Action
        submit_button.click(
            ask_question,
            inputs=[
                question,
                model,
                api_key,
                semantic_usage,
                initial_top_n,
                final_top_n,
            ],
            outputs=[answer, chunks_response],
        )

        # Footer
        gr.Markdown(
            """
            <h3 style='text-align: center; color: #9d00ff;'>Made by Maksym Batiuk and Olena Morozevych.</h3>
            """
        )

    # Launch the Gradio app
    demo.launch(server_name="0.0.0.0", server_port=7860)