Turn a data scientist's notebook into a production-ready service

Turn a data scientist's notebook into a production-ready service | Buncho

[1] df = pd.read_csv("C:/Users/anita/Desktop/churn_final_v3.csv"); api_key = "sk-..."   # path and a key in plain text
[2] df = df.fillna(df.mean(numeric_only=True))                    # mean over ALL rows
[3] df = pd.get_dummies(df, columns=["plan", "city"])             # on the whole frame
[4] X = StandardScaler().fit_transform(df.drop("churned", axis=1))  # scaler not saved
[5] X_train, X_test, y_train, y_test = train_test_split(X, df.churned, test_size=0.2)   # random split
[6] model = GradientBoostingClassifier().fit(X_train, y_train)    # no random_state
[7] print(roc_auc_score(y_test, model.predict_proba(X_test)[:, 1]))   # 0.93
[8] pickle.dump(model, open("model.pkl", "wb"))                   # model only

data: 18 months of daily customer snapshots; many rows per customer
requirements.txt: not present   |   tests: none

Requirements from the business

{
  "deliverables": ["daily batch scores for all active customers", "REST endpoint for single-customer scoring"],
  "deadline_weeks": 3,
  "latency_target_ms_p95": 300,
  "batch_size": "about 2 million customers per day",
  "constraint": "retention team only acts on the top 5% of risk scores",
  "notes": "marketing wants a go-live date this month; model owner is on leave in week 2"
}

Turn a data scientist's notebook into a production-ready service

Scenario data

How to answer

Hints

AI feedback

Community solutions