app.py

import gradio as gr
import requests
import json
import speech_recognition as sr

GEMINI_API_KEY = "AIzaSyAZ3WSpx_o53zpmhIJRzR0JMsiBOTnttbg"
API_URL = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent?key={GEMINI_API_KEY}"

INTRO_TEXT = """
<div>
  <h1>HydroChat – Your Personalized Assistant on Molecular Hydrogen Supplements</h1>
  <p>This chatbot delivers evidence‑based guidance on using molecular hydrogen (H<sub>2</sub>) supplements in conjunction with conventional medicine and everyday health practices.</p>
  <p>Our mission is to translate peer‑reviewed clinical research into practical advice so that users can make informed decisions about hydrogen‑rich coral‑calcium capsules.</p>
</div>
"""


def call_gemini(message: str) -> str:
    """Send the user message as a structured prompt to the Gemini endpoint and
    return the assistant's reply text. Errors are surfaced in‑line for easier debugging.
    """
    headers = {"Content-Type": "application/json"}

    prompt = f"""You are HydroChat, an evidence‑based clinical assistant for an oral capsule product that contains\n- Coral calcium as a stable carrier of molecular hydrogen (H₂)\n- Additional vitamins and nutrients\n\nYour task is to provide scientifically sound health education and supplement‑drug interaction advice. Follow these principles:\n1. Encourage users to explore **synergistic use** of this hydrogen capsule with existing supplements or prescription medications, while reminding them to consult a qualified healthcare professional.\n2. Keep the discussion anchored on **molecular hydrogen**: its documented antioxidative, anti‑inflammatory, and cytoprotective mechanisms, and coral calcium's role as a solid‑state hydrogen carrier.\n3. All answers must be **concise, evidence‑based, and free of exaggerated or speculative claims**.\n4. Respond in the **same language** the user used.\n\nUser message: {message}\n"""

    data = {"contents": [{"parts": [{"text": prompt}]}]}

    response = requests.post(API_URL, headers=headers, data=json.dumps(data))
    if response.status_code == 200:
        try:
            return response.json()["candidates"][0]["content"]["parts"][0]["text"]
        except (KeyError, IndexError) as err:
            return f"[Parsing Error]: {err}"
    else:
        return f"[API Error {response.status_code}]: {response.text}"


def transcribe_audio(audio_file: str, lang_code: str) -> str:
    """Transcribe an uploaded audio file using Google Speech Recognition."""
    recognizer = sr.Recognizer()
    with sr.AudioFile(audio_file) as source:
        audio = recognizer.record(source)
    try:
        return recognizer.recognize_google(audio, language=lang_code)
    except Exception as err:
        return f"[Speech‑to‑Text Error]: {err}"


with gr.Blocks(title="HydroChat – Molecular Hydrogen Supplement Assistant") as demo:
    gr.HTML(INTRO_TEXT)
    chatbot = gr.Chatbot(height=400, label="HydroChat Conversation", show_copy_button=True)

    with gr.Row():
        msg = gr.Textbox(label="Enter your question", placeholder="e.g. I am currently on __ therapy; can I take __?", scale=6)
        lang_select = gr.Dropdown(
            label="Speech recognition language", choices=[
                ("中文 (台灣)", "zh-TW"),
                ("English (US)", "en-US"),
                ("日本語", "ja-JP"),
                ("한국어", "ko-KR"),
                ("Bahasa Indonesia", "id-ID"),
                ("Tiếng Việt", "vi-VN"),
                ("Français", "fr-FR"),
                ("Deutsch", "de-DE"),
            ], value="en-US", scale=2
        )

    with gr.Row():
        audio_input = gr.Audio(label="🎙️ Record voice", type="filepath")
        voice_to_text = gr.Button("Transcribe 🎤")

    with gr.Row():
        ask = gr.Button("Submit")
        clear = gr.Button("Clear Chat")

    # ----- Callbacks -------------------------------------------------------
    def respond(message: str, history: list):
        if not message.strip():
            return "", history
        reply = call_gemini(message)
        history.append({"role": "user", "content": message})
        history.append({"role": "assistant", "content": reply})
        return "", history

    def handle_audio(audio_path: str, lang_code: str):
        if not audio_path:
            return ""
        return transcribe_audio(audio_path, lang_code)

    msg.submit(respond, inputs=[msg, chatbot], outputs=[msg, chatbot])
    ask.click(respond, inputs=[msg, chatbot], outputs=[msg, chatbot])
    clear.click(lambda: [], outputs=chatbot)
    voice_to_text.click(handle_audio, inputs=[audio_input, lang_select], outputs=msg)


demo.launch()