--- a
+++ b/prediction_api/src/prediction_system.py
@@ -0,0 +1,60 @@
+import logging
+from os import getenv, path
+
+from dotenv import load_dotenv
+import numpy as np
+from sklearn.preprocessing import MinMaxScaler
+from tensorflow.keras.models import load_model
+
+load_dotenv()
+
+logging.basicConfig(
+    level=logging.DEBUG,
+    format="%(asctime)s [%(levelname)s]: %(message)s",
+    handlers=[
+        logging.StreamHandler()
+    ],
+)
+
+logger = logging.getLogger(__name__)
+
+logger.info('Loading prediction system')
+THRESHOLD = float(getenv('THRESHOLD', '0.7'))
+models_folder = getenv('MODELS_FOLDER', '')
+model_name = getenv('MODEL_NAME', 'autoposture-model.h5')
+model = None
+
+def preprocess_sequences(sequences):
+    sequences = np.array(sequences)
+    scaler = MinMaxScaler()
+    normalized_sequences = np.zeros_like(sequences)
+    for i in range(sequences.shape[0]):
+        for j in range(sequences.shape[1]):
+            # Flatten, Normalize, Reshape
+            landmarks_flattened = np.reshape(sequences[i, j], (-1, 1))
+            landmarks_normalized = scaler.fit_transform(landmarks_flattened)
+            normalized_landmarks = np.reshape(landmarks_normalized, sequences[i, j].shape)
+            # Save process in empty ndarray
+            normalized_sequences[i, j] = normalized_landmarks
+
+    return normalized_sequences
+
+
+def get_prediction(payload : dict):
+    received_array = np.array(payload['array'])
+    sequence = preprocess_sequences(received_array)
+
+    prediction = model.predict(sequence, verbose=0).tolist()
+    score = prediction[0][0]
+    response = { 'score': score }
+    return response
+
+def initialize_model():
+    global model
+    model_path = path.join(models_folder, model_name)
+    model = load_model(model_path)
+    logger.info(model.summary())
+
+def healthcheck():
+    global model
+    return True if model else False