app.py

import re
import gradio as gr

from prompts import SYSTEM_PROMPT, PHASE_SWITCH_REMINDER
from parse_feedback import parse_feedback, render_feedback_table, strip_markdown
from stt_engine import transcribe
from llm_engine import chat as llm_chat
from tts_engine import synthesize

TERMINATE_RE = re.compile(r"(fin\s+de\s+(la\s+)?séance|session\s+terminée)", re.IGNORECASE)

# 7 outputs: chatbot, audio_output, state, feedback_intro, feedback_table, feedback_panel, status

def _idle_feedback():
    return "", [], gr.update(open=False)

def _show_feedback(state, clean):
    entries = parse_feedback(clean)
    table = render_feedback_table(entries) if entries else []
    intro = clean
    if "Disse:" in intro:
        intro = intro.split("Disse:")[0].strip()
    return intro, table, gr.update(open=True)

def _chat_val(state):
    return state["messages"]

def _make_audio(audio_bytes):
    return audio_bytes if audio_bytes else None

def process_turn(audio_path, state):
    state = dict(state)
    if not audio_path:
        yield _chat_val(state), None, state, *_idle_feedback(), ""
        return

    # 1. STT
    yield _chat_val(state), None, state, *_idle_feedback(), "🎙 Transcription…"
    user_text = transcribe(audio_path)
    if not user_text or len(user_text.strip()) < 2:
        yield _chat_val(state), None, state, *_idle_feedback(), "⛔ Parlez plus fort ou plus longtemps."
        return

    state["messages"].append({"role": "user", "content": user_text.strip()})

    if TERMINATE_RE.search(user_text):
        yield from _end_session(state)
        return

    # 2. LLM
    yield _chat_val(state), None, state, *_idle_feedback(), "🧠 Réflexion…"
    response = llm_chat(state["messages"])
    if not response:
        yield _chat_val(state), None, state, *_idle_feedback(), "⛔ Erreur du modèle. Réessayez."
        return

    clean = strip_markdown(response)
    state["messages"].append({"role": "assistant", "content": clean})

    # 3. TTS
    yield _chat_val(state), None, state, *_idle_feedback(), "🔊 Synthèse vocale…"
    audio_bytes = synthesize(clean)

    yield _chat_val(state), _make_audio(audio_bytes), state, *_idle_feedback(), ""


def _end_session(state):
    state["messages"].append({"role": "user", "content": PHASE_SWITCH_REMINDER})

    yield _chat_val(state), None, state, *_idle_feedback(), "📝 Génération du récapitulatif…"
    response = llm_chat(state["messages"])
    if not response:
        yield _chat_val(state), None, state, *_idle_feedback(), "⛔ Erreur lors de la génération du bilan."
        return

    clean = strip_markdown(response)
    state["messages"].append({"role": "assistant", "content": clean})
    state["phase"] = 2

    intro, table, accordion = _show_feedback(state, clean)
    audio_bytes = synthesize(intro)

    yield _chat_val(state), _make_audio(audio_bytes), state, intro, table, accordion, ""


def end_session_click(state):
    state = dict(state)
    yield from _end_session(state)


def reset_session():
    state = {"messages": [], "phase": 1, "turn_count": 0}
    state["messages"].append({"role": "system", "content": SYSTEM_PROMPT})
    return [], None, state, *_idle_feedback(), ""


# ---- Init state ----
initial_messages = [{"role": "system", "content": SYSTEM_PROMPT}]
initial_state = {"messages": list(initial_messages), "phase": 1, "turn_count": 0}

# ---- Gradio UI ----
custom_css = open("style.css", encoding="utf-8").read()

with gr.Blocks(
    css=custom_css,
    title="Patient Virtuel · Hygiéniste Pro",
    theme=gr.themes.Soft(primary_hue="orange"),
) as demo:
    gr.HTML('<div class="atmosphere"></div>')

    gr.Markdown(
        '<h1 class="app-title" style="text-align:center; font-weight:400; '
        'font-family:Cormorant Garamond,serif; color:white; margin-bottom:0; '
        'font-size:28px; letter-spacing:0.02em;">'
        "Patient Virtuel · Hygiéniste Pro</h1>"
    )

    state = gr.State(initial_state)

    with gr.Row():
        with gr.Column(scale=1, min_width=280):
            audio_input = gr.Audio(
                sources=["microphone"],
                type="filepath",
                show_label=False,
                show_download_button=False,
                waveform_options={"waveform_color": "#ff4e00", "show_controls": False},
            )

            gr.Markdown(
                '<p style="font-size:13px; color:#888; text-align:center; '
                'margin-top:4px;">Appuyez pour parler, relâchez pour envoyer</p>'
            )

            with gr.Row():
                btn_end = gr.Button("🟠 Terminer la séance", variant="stop", scale=2)
                btn_clear = gr.Button("🗑 Nouvelle", variant="secondary", scale=1)

            status = gr.Markdown("", elem_id="status-bar")

        with gr.Column(scale=2):
            chatbot = gr.Chatbot(
                value=list(initial_messages),
                type="messages",
                label="Conversation",
                height=480,
                avatar_images=(None, "🤖"),
                show_copy_button=False,
                sanitize_html=True,
                render_markdown=False,
            )

            audio_output = gr.Audio(
                label="Réponse audio",
                autoplay=True,
                show_download_button=False,
                interactive=False,
                waveform_options={"waveform_color": "#ff4e00", "show_controls": False},
            )

    with gr.Row():
        feedback_panel = gr.Accordion(
            label="📋 Récapitulatif de la séance",
            open=False,
        )
        with feedback_panel:
            feedback_intro = gr.Markdown("")
            feedback_table = gr.Dataframe(
                headers=["Disse", "Correction", "Explication"],
                datatype=["str", "str", "str"],
                wrap=True,
                interactive=False,
                label="Erreurs relevées",
                show_label=False,
            )

    gr.Markdown(
        '<p style="font-size:11px; color:#555; text-align:center; margin-top:16px;">'
        "License CC-BY-NC 4.0 (Voxtral TTS) — démonstration non-commerciale. "
        "Modèle: Qwen/Qwen3.6-27B, STT: faster-whisper.</p>"
    )

    # ---- Event wiring ----
    outputs = [chatbot, audio_output, state, feedback_intro, feedback_table, feedback_panel, status]

    audio_input.change(fn=process_turn, inputs=[audio_input, state], outputs=outputs)
    btn_end.click(fn=end_session_click, inputs=[state], outputs=outputs)
    btn_clear.click(fn=reset_session, inputs=[], outputs=outputs)

# ---- Launch ----
if __name__ == "__main__":
    demo.launch(show_api=False)