app.py

import os
import time
from typing import List, Tuple, Optional

import google.generativeai as genai
import gradio as gr
from PIL import Image


# API KEY (ENVIRONMENT VARIABLE ONLY – NO HARDCODING)

GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY") or "AIzaSyAx65ry4WsiWdd9Ixh-jsPfx4Lg20NZ5sY"
if not GOOGLE_API_KEY:
    raise ValueError("GOOGLE_API_KEY is not set. Add it as an environment variable or HF Secret.")

genai.configure(api_key=GOOGLE_API_KEY)


# MODEL CONFIGURATION

IMAGE_WIDTH = 512

system_instruction_analysis = (
    "You are an expert of the given topic. Analyze the provided text with a focus on the topic, "
    "identifying recent issues, recent insights, or improvements relevant to academic standards "
    "and effectiveness. Offer actionable advice for enhancing knowledge and suggest real-life examples."
)

#  Supported public multimodal model
MODEL_NAME = "gemini-3-flash-preview"

model = genai.GenerativeModel(
    MODEL_NAME,
    system_instruction=system_instruction_analysis
)


# HELPER FUNCTIONS

def preprocess_stop_sequences(stop_sequences: str) -> Optional[List[str]]:
    return [seq.strip() for seq in stop_sequences.split(",")] if stop_sequences else None


def preprocess_image(image: Image.Image) -> Image.Image:
    image_height = int(image.height * IMAGE_WIDTH / image.width)
    return image.resize((IMAGE_WIDTH, image_height))


def user(text_prompt: str, chatbot: List[Tuple[str, str]]):
    return "", chatbot + [[text_prompt, None]]


def bot(
    image_prompt: Optional[Image.Image],
    temperature: float,
    max_output_tokens: int,
    stop_sequences: str,
    top_k: int,
    top_p: float,
    chatbot: List[Tuple[str, str]]
):
    text_prompt = chatbot[-1][0].strip() if chatbot[-1][0] else None

    if not text_prompt and not image_prompt:
        chatbot[-1][1] = "Prompt cannot be empty. Please enter text or upload an image."
        yield chatbot
        return

    if image_prompt and not text_prompt:
        text_prompt = "Describe the image."
    elif image_prompt and text_prompt:
        text_prompt = f"{text_prompt}. Also analyze the provided image."

    generation_config = genai.types.GenerationConfig(
        temperature=temperature,
        max_output_tokens=max_output_tokens,
        stop_sequences=preprocess_stop_sequences(stop_sequences),
        top_k=top_k,
        top_p=top_p,
    )

    inputs = (
        [text_prompt]
        if image_prompt is None
        else [text_prompt, preprocess_image(image_prompt)]
    )

    try:
        response = model.generate_content(
            inputs,
            generation_config=generation_config
        )
        chatbot[-1][1] = response.text
        yield chatbot

    except Exception as e:
        chatbot[-1][1] = f" Error: {str(e)}"
        yield chatbot



# UI COMPONENTS

chatbot_component = gr.Chatbot(label="Chatbot", bubble_full_width=False)

text_prompt_component = gr.Textbox(
    label="Ask",
    placeholder="Type your question here...",
    lines=3
)

image_prompt_component = gr.Image(
    type="pil",
    label="Input Image (Optional: Figure / Graph)"
)

run_button_component = gr.Button("Submit")

temperature_component = gr.Slider(
    0, 1.0, 0.4, 0.05, label="Creativity (Temperature)"
)

max_output_tokens_component = gr.Slider(
    1, 2048, 1024, 1, label="Response Length (Token Limit)"
)

stop_sequences_component = gr.Textbox(
    label="Stop Sequences (Optional)",
    placeholder="STOP, END"
)

top_k_component = gr.Slider(
    1, 40, 32, 1, label="Top-K Sampling"
)

top_p_component = gr.Slider(
    0, 1, 1, 0.01, label="Top-P Sampling"
)

example_scenarios = [
    "Describe Multimodal AI",
    "What is the difference between multi-agent LLMs and multi-agent systems?",
    "Why is it difficult to integrate multimodality in prompts?"
]

example_images = [["ex1.png"], ["ex2.png"]]


# GRADIO APP

with gr.Blocks(theme="earneleh/paris") as demo:
    gr.Markdown(
        "<h1 style='font-size:36px; font-weight:bold;'>Google Gemini</h1>"
    )

    chatbot_component.render()

    with gr.Row():
        text_prompt_component.render()
        image_prompt_component.render()
        run_button_component.render()

    with gr.Accordion("🧪 Example Text 💬", open=False):
        example_radio = gr.Radio(
            choices=example_scenarios,
            label="Example Queries"
        )
        example_radio.change(
            fn=lambda q: q,
            inputs=[example_radio],
            outputs=[text_prompt_component]
        )

    with gr.Accordion("🧪 Example Image 🩻", open=False):
        gr.Examples(
            examples=example_images,
            inputs=[image_prompt_component],
            label="Example Figures"
        )

    with gr.Accordion("🛠️ Customize", open=False):
        temperature_component.render()
        max_output_tokens_component.render()
        stop_sequences_component.render()
        top_k_component.render()
        top_p_component.render()

    run_button_component.click(
        fn=user,
        inputs=[text_prompt_component, chatbot_component],
        outputs=[text_prompt_component, chatbot_component]
    ).then(
        fn=bot,
        inputs=[
            image_prompt_component,
            temperature_component,
            max_output_tokens_component,
            stop_sequences_component,
            top_k_component,
            top_p_component,
            chatbot_component,
        ],
        outputs=[chatbot_component]
    )

demo.launch(share=True, debug=True)