app.py

import os, json, requests, threading
from flask import Flask, request, jsonify
from groq_transcribe import groq_transcribe
from labelstudio_inject import upload_prediction
from correction_collector import save_correction
from dataset_collector import prepare_dataset

app = Flask(__name__)

# Load env
LABEL_URL = os.environ.get("LABEL_STUDIO_URL")
API_TOKEN = os.environ.get("LABEL_STUDIO_API_TOKEN")
HEADERS = {"Authorization": f"Token {API_TOKEN}", "Content-Type": "application/json"}

def get_latest_project_id():
    """Dynamic Project ID fetch"""
    r = requests.get(f"{LABEL_URL}/api/projects/", headers=HEADERS)
    r.raise_for_status()
    projects = r.json()
    latest_project = sorted(projects, key=lambda x: x['id'], reverse=True)[0]
    return latest_project['id']

PROJECT_ID = get_latest_project_id()

# -----------------------------
# Routes
# -----------------------------

@app.route("/ml_health", methods=["GET"])
def ml_health():
    return jsonify({
        "engine": "groq",
        "label_studio_url": LABEL_URL,
        "language": os.environ.get("WHISPER_LANGUAGE"),
        "model_version": os.environ.get("GROQ_MODEL"),
        "project_id": PROJECT_ID,
        "status": "ok"
    })

@app.route("/auto_transcribe", methods=["POST"])
def auto_transcribe():
    """Upload audio and auto transcribe using Groq, inject to Label Studio"""
    audio_file = request.files.get("audio")
    if not audio_file:
        return jsonify({"error": "No audio file uploaded"}), 400

    audio_path = f"/tmp/{audio_file.filename}"
    audio_file.save(audio_path)

    # Transcription in background thread
    def transcribe_task(path):
        try:
            transcript = groq_transcribe(path)
            # For demo, pick first task in project
            task_id = request.form.get("task_id")
            if not task_id:
                # fallback: get latest task from Label Studio
                r = requests.get(f"{LABEL_URL}/api/projects/{PROJECT_ID}/tasks/", headers=HEADERS)
                r.raise_for_status()
                tasks = r.json()
                if tasks:
                    task_id = tasks[0]['id']
            if task_id:
                upload_prediction(task_id, transcript)
            print(f"Transcribed {path} → Label Studio task {task_id}")
        except Exception as e:
            print("Error in transcription:", e)

    threading.Thread(target=transcribe_task, args=(audio_path,)).start()
    return jsonify({"status": "started"})

@app.route("/save_correction", methods=["POST"])
def save_correction_route():
    """Human corrected transcription"""
    data = request.json
    task_id = data.get("task_id")
    human_text = data.get("human_text")
    groq_text = data.get("groq_text")
    audio_path = data.get("audio_path")
    if not all([task_id, human_text, groq_text, audio_path]):
        return jsonify({"error": "Missing field"}), 400

    save_correction(task_id, human_text, groq_text, audio_path)
    return jsonify({"status": "saved"})

@app.route("/prepare_dataset", methods=["POST"])
def prepare_dataset_route():
    """Prepare dataset for fine-tune"""
    prepare_dataset()
    return jsonify({"status": "dataset_ready"})

# -----------------------------
# Run App
# -----------------------------
if __name__ == "__main__":
    port = int(os.environ.get("PORT", 7860))
    app.run(host="0.0.0.0", port=port)