Reference: A Production ML Pipeline

fraud-ml/
  pipelines/train.py       # scheduled training job
  pipelines/features.py    # feature store writes
  serving/app.py           # FastAPI inference
  models/                  # exported weights (S3 in prod)
  eval/metrics.py          # offline gates
  monitoring/drift.py      # prod distribution checks

"""pipelines/train.py - batch training with eval gate."""
from __future__ import annotations
 
import json
from pathlib import Path
 
import joblib
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.metrics import roc_auc_score
from sklearn.model_selection import train_test_split
 
from pipelines.features import load_feature_frame
 
 
def train(version: str, min_auc: float = 0.82) -> Path:
    X, y = load_feature_frame()
    X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, random_state=42)
    model = GradientBoostingClassifier(random_state=42)
    model.fit(X_train, y_train)
    auc = roc_auc_score(y_test, model.predict_proba(X_test)[:, 1])
    if auc < min_auc:
        raise RuntimeError(f"AUC {auc:.3f} below gate {min_auc}")
    out = Path(f"models/{version}/model.joblib")
    out.parent.mkdir(parents=True, exist_ok=True)
    joblib.dump(model, out)
    out.with_suffix(".metrics.json").write_text(json.dumps({"auc": auc}))
    return out

"""serving/app.py - inference API."""
from __future__ import annotations
 
import joblib
from fastapi import FastAPI
from pydantic import BaseModel, Field
 
app = FastAPI()
MODEL = joblib.load("models/prod/model.joblib")
 
 
class Features(BaseModel):
    amount: float = Field(gt=0)
    merchant_risk_score: float
 
 
class Score(BaseModel):
    fraud_probability: float
    model_version: str = "2026.07.09"
 
 
@app.post("/v1/score", response_model=Score)
def score(body: Features) -> Score:
    prob = float(MODEL.predict_proba([[body.amount, body.merchant_risk_score]])[0, 1])
    return Score(fraud_probability=prob)

# monitoring/drift.py - simplified PSI check
def population_stability_index(expected: list[float], actual: list[float]) -> float:
    ...