diff --git a/man/supervisord.conf b/man/supervisord.conf
index 7b02a24..03bc5ca 100644
--- a/man/supervisord.conf
+++ b/man/supervisord.conf
@@ -13,7 +13,7 @@ user=root                                       ; default user
 file = /tmp/supervisor.sock                     ; supervisor unix http server sock file
 
 [program:microservicios]
-command=python3 flask.py
+command=python3 app.py
 directory=/app
 stdout_logfile=/dev/fd/1
 stdout_logfile_maxbytes=0
diff --git a/src/app.py b/src/app.py
new file mode 100644
index 0000000..18a0d5d
--- /dev/null
+++ b/src/app.py
@@ -0,0 +1,94 @@
+from flask import Flask, jsonify, request
+import joblib
+import os
+from datetime import date
+
+from db import fetch_data
+import threading
+from model_registry import register, load_meta
+from models_train import train_demand_model, train_nulo_model
+
+app = Flask(__name__)
+MODEL_DIR = "models"
+
+def load_model(model_type):
+    meta = load_meta()
+    file = meta["current"].get(model_type)
+    if not file:
+        return None
+    return joblib.load(os.path.join(MODEL_DIR, file))
+
+def background_train():
+    print("Fetching data...")
+    df = fetch_data()
+    print(f"Data fetched: {len(df)} rows")
+    
+    today = date.today().isoformat()
+    os.makedirs(MODEL_DIR, exist_ok=True)
+
+    print("Training demand model...")
+    demand_model = train_demand_model(df)
+    demand_file = f"demand_xgb_{today}.joblib"
+    joblib.dump(demand_model, f"{MODEL_DIR}/{demand_file}")
+    register("demand", demand_file, len(df))
+    print("Demand model trained and saved.")
+
+    print("Training nulo model...")
+    nulo_model = train_nulo_model(df)
+    nulo_file = f"nulo_xgb_{today}.joblib"
+    joblib.dump(nulo_model, f"{MODEL_DIR}/{nulo_file}")
+    register("nulo", nulo_file, len(df))
+    print("Nulo model trained and saved.")
+
+@app.route("/train", methods=["POST"])
+def train():
+    threading.Thread(target=background_train).start()
+    return jsonify({"status": "training started"})
+
+@app.route("/predict_demand", methods=["GET"])
+def predict_demand():
+    h3 = int(request.args["h3"])
+    week = int(request.args["week"])
+    dow = int(request.args["dow"])
+    hour = int(request.args["hour"])
+
+    model = load_model("demand")
+    X = [[h3, week, dow, hour]]
+    pred = model.predict(X)[0]
+
+    return jsonify({"expected_demand": float(pred)})
+
+@app.route("/predict_nulo", methods=["GET"])
+def predict_nulo():
+    h3 = int(request.args["h3"])
+    week = int(request.args["week"])
+    dow = int(request.args["dow"])
+    hour = int(request.args["hour"])
+
+    model = load_model("nulo")
+    X = [[h3, week, dow, hour]]
+    prob = model.predict_proba(X)[0][1]
+
+    return jsonify({"nulo_probability": float(prob)})
+
+@app.route("/predict", methods=["GET"])
+def predict_legacy():
+    h3 = int(request.args["h3"])
+    week = int(request.args["week"])
+    dow = int(request.args["dow"])
+    hour = int(request.args["hour"])
+
+    model = load_model("nulo")
+    X = [[h3, week, dow, hour]]
+    prob = model.predict_proba(X)[0][1]
+
+    return jsonify({
+        "predicted_nulo_prob": float(prob)
+    })
+
+@app.route("/models", methods=["GET"])
+def models():
+    return jsonify(load_meta())
+
+if __name__ == "__main__":
+    app.run(host="0.0.0.0", port=5000)
diff --git a/src/db.py b/src/db.py
new file mode 100644
index 0000000..3c42442
--- /dev/null
+++ b/src/db.py
@@ -0,0 +1,37 @@
+from sqlalchemy import create_engine, text
+from sqlalchemy.exc import SQLAlchemyError
+import pandas as pd
+import os
+
+# Configuración de la conexión (puedes usar variables de entorno)
+DB_USER = os.getenv("DB_USER", "postgres")
+DB_PASSWORD = os.getenv("DB_PASSWORD", "tfmuocdfcarvajal")
+DB_HOST = os.getenv("DB_HOST", "10.10.5.32")
+DB_PORT = os.getenv("DB_PORT", "5432")
+DB_NAME = os.getenv("DB_NAME", "postgres")
+
+DATABASE_URL = f"postgresql+psycopg2://{DB_USER}:{DB_PASSWORD}@{DB_HOST}:{DB_PORT}/{DB_NAME}"
+
+# Crear el engine de SQLAlchemy
+engine = create_engine(DATABASE_URL, echo=False, future=True)
+
+
+def fetch_data():
+    query = """
+        SELECT
+          h3,
+          week,
+          dow,
+          hour,
+          total_events,
+          total_nulos,
+          nulo_rate
+        FROM demanda_h3_hour_ml
+    """
+    try:
+        with engine.connect() as conn:
+            df = pd.read_sql(text(query), conn)
+        return df
+    except SQLAlchemyError as e:
+        print(f"Error al ejecutar la consulta: {e}")
+        return pd.DataFrame()
\ No newline at end of file
diff --git a/src/db_warning.py b/src/db_warning.py
new file mode 100644
index 0000000..663cd67
--- /dev/null
+++ b/src/db_warning.py
@@ -0,0 +1,28 @@
+import psycopg2
+import pandas as pd
+import os
+
+DB_CONFIG = {
+    'dbname': os.getenv('DB_NAME', 'postgres'),
+    'user': os.getenv('DB_USER', 'postgres'),
+    'password': os.getenv('DB_PASSWORD', 'tfmuocdfcarvajal'),
+    'host': os.getenv('DB_HOST', '10.10.5.32'),
+    'port': os.getenv('DB_PORT', '5432')
+}
+
+def fetch_data():
+    conn = psycopg2.connect(**DB_CONFIG)
+    query = """
+        SELECT
+          h3,
+          week,
+          dow,
+          hour,
+          total_events,
+          total_nulos,
+          nulo_rate
+        FROM demanda_h3_hour_ml
+    """
+    df = pd.read_sql(query, conn)
+    conn.close()
+    return df
\ No newline at end of file
diff --git a/src/flask.py b/src/flask.py
deleted file mode 100644
index ca0b0a5..0000000
--- a/src/flask.py
+++ /dev/null
@@ -1,85 +0,0 @@
-from flask import Flask, jsonify, request
-import os
-import psycopg2
-import pandas as pd
-import xgboost as xgb
-import joblib
-
-app = Flask(__name__)
-
-# Configuración de la base de datos
-DB_CONFIG = {
-    'dbname': os.getenv('DB_NAME', 'postgres'),
-    'user': os.getenv('DB_USER', 'postgres'),
-    'password': os.getenv('DB_PASSWORD', 'tfmuocdfcarvajal'),
-    'host': os.getenv('DB_HOST', '10.10.5.32'),
-    'port': os.getenv('DB_PORT', '5432')
-}
-
-# Ruta para entrenar o cargar el modelo
-MODEL_PATH = 'modelo_xgb.joblib'
-
-def get_db_connection():
-    conn = psycopg2.connect(**DB_CONFIG)
-    return conn
-
-def fetch_data():
-    conn = get_db_connection()
-    query = """
-        SELECT h3, hour, dow, total_events, total_nulos, nulo_rate
-        FROM demanda_h3_hour
-    """
-    df = pd.read_sql(query, conn)
-    conn.close()
-    return df
-
-def train_model(df):
-    # Variables predictoras
-    X = df[['hour', 'dow', 'total_events']]
-    # Variable objetivo: nulo_rate > 0 -> 1, else 0
-    y = (df['nulo_rate'] > 0).astype(int)
-    
-    model = xgb.XGBClassifier(
-        max_depth=4, n_estimators=100, learning_rate=0.1, use_label_encoder=False, eval_metric='logloss'
-    )
-    model.fit(X, y)
-    joblib.dump(model, MODEL_PATH)
-    return model
-
-def load_model():
-    try:
-        model = joblib.load(MODEL_PATH)
-        return model
-    except:
-        df = fetch_data()
-        return train_model(df)
-
-model = load_model()
-
-@app.route('/predict', methods=['GET'])
-def predict():
-    # Parámetros de la consulta
-    hour = int(request.args.get('hour'))
-    dow = int(request.args.get('dow'))
-    total_events = int(request.args.get('total_events', 1))  # valor por defecto si no se pasa
-    
-    X_pred = pd.DataFrame([[hour, dow, total_events]], columns=['hour', 'dow', 'total_events'])
-    prob = model.predict_proba(X_pred)[0][1]  # Probabilidad de nulo
-    return jsonify({
-        'hour': hour,
-        'dow': dow,
-        'total_events': total_events,
-        'predicted_nulo_prob': float(prob)
-    })
-
-@app.route('/demand', methods=['GET'])
-def demand():
-    df = fetch_data()
-    X_pred = df[['hour', 'dow', 'total_events']]
-    df['predicted_nulo_prob'] = model.predict_proba(X_pred)[:,1].astype(float)
-    # Convertimos a JSON
-    result = df.to_dict(orient='records')
-    return jsonify(result)
-
-if __name__ == '__main__':
-    app.run(host='0.0.0.0', port=5000)
\ No newline at end of file
diff --git a/src/model_registry.py b/src/model_registry.py
new file mode 100644
index 0000000..6fa0f06
--- /dev/null
+++ b/src/model_registry.py
@@ -0,0 +1,30 @@
+import json
+import os
+from datetime import datetime
+
+MODEL_DIR = "models"
+META_FILE = os.path.join(MODEL_DIR, "metadata.json")
+
+def load_meta():
+    if not os.path.exists(META_FILE):
+        return {"current": {}, "history": []}
+    with open(META_FILE) as f:
+        return json.load(f)
+
+def save_meta(meta):
+    os.makedirs(MODEL_DIR, exist_ok=True)
+    with open(META_FILE, "w") as f:
+        json.dump(meta, f, indent=2)
+
+def register(model_type, file_name, rows):
+    meta = load_meta()
+    entry = {
+        "type": model_type,
+        "file": file_name,
+        "trained_at": datetime.utcnow().isoformat() + "Z",
+        "rows": rows
+    }
+    meta["history"].append(entry)
+    meta["current"][model_type] = file_name
+    save_meta(meta)
+    return entry
diff --git a/src/models_train.py b/src/models_train.py
new file mode 100644
index 0000000..509ed19
--- /dev/null
+++ b/src/models_train.py
@@ -0,0 +1,33 @@
+import xgboost as xgb
+
+def train_demand_model(df):
+    df["h3_int"] = df["h3"].apply(lambda x: int(x, 16))
+
+    X = df[["h3_int", "week", "dow", "hour"]]
+    y = df["total_events"]
+
+    model = xgb.XGBRegressor(
+        n_estimators=300,
+        max_depth=8,
+        learning_rate=0.05,
+        objective="reg:squarederror"
+    )
+
+    model.fit(X, y)
+    return model
+
+def train_nulo_model(df):
+    df["h3_int"] = df["h3"].apply(lambda x: int(x, 16))
+
+    X = df[["h3_int", "week", "dow", "hour"]]
+    y = (df["nulo_rate"] > 0).astype(int)
+
+    model = xgb.XGBClassifier(
+        n_estimators=200,
+        max_depth=5,
+        learning_rate=0.1,
+        eval_metric="logloss"
+    )
+
+    model.fit(X, y)
+    return model
diff --git a/src/pyproject.toml b/src/pyproject.toml
index 73f8372..3287afb 100644
--- a/src/pyproject.toml
+++ b/src/pyproject.toml
@@ -16,6 +16,7 @@ authors = [
 dependencies = [
    'flask',
    'psycopg2-binary',
+   'sqlalchemy',
    'pandas',
    'xgboost',
    'scikit-learn',
@@ -24,7 +25,7 @@ dependencies = [
 
 [tool.setuptools]
 include-package-data = true
-py-modules = ["flask"]
+py-modules = ["app", "db", "model_registry", "models_train"]
 
 [tool.setuptools.packages]
 find = {}