diff --git a/tools/recommend.py b/tools/recommend.py
new file mode 100644
index 00000000..d8d1f723
--- /dev/null
+++ b/tools/recommend.py
@@ -0,0 +1,50 @@
+"""
+This is a music recommendation system for maimai2 using implicit ALS.
+"""
+import json
+from pathlib import Path
+
+import pandas as pd
+import requests
+import scipy.sparse as sp
+import implicit
+from hypy_utils.logging_utils import setup_logger
+
+BASE_URL = "https://aquadx.net/aqua/api/v2/game"
+GAME = "mai2"
+BOT_SECRET = "meow"
+
+log = setup_logger()
+
+
+if __name__ == '__main__':
+    # Load the CSV data
+    log.info("Loading data...")
+    # data = pd.read_csv("data.csv")
+    resp = requests.get(f"{BASE_URL}/{GAME}/recommender-fetch", params={"botSecret": BOT_SECRET})
+    assert resp.status_code == 200, f"Failed to fetch data: {resp.status_code} {resp.text}"
+    data = pd.read_csv(resp.text)
+
+    # Create a user-item matrix
+    log.info("Creating user-item matrix...")
+    user_item_matrix = sp.csr_matrix((
+        data['count'],
+        (data['user_id'], data['music_id'])
+    ))
+
+    # Train an ALS model
+    log.info("Training ALS model...")
+    model = implicit.als.AlternatingLeastSquares(factors=50, regularization=0.01, iterations=15)
+    model.fit(user_item_matrix)
+
+    # Generate recommendations for each user
+    log.info("Generating recommendations...")
+    recommendations = {}
+    for user_id in range(user_item_matrix.shape[0]):  # Loop over all users
+        rec, prob = model.recommend(user_id, user_item_matrix[user_id], N=20)
+        recommendations[user_id] = [int(item) for item in rec]
+
+    # Save recommendations to a file
+    log.info("Saving recommendations...")
+    # Path("recommendations.json").write_text(json.dumps(recommendations))
+    resp = requests.post(f"{BASE_URL}/{GAME}/recommender-update", params={"botSecret": BOT_SECRET}, json=recommendations)