diff --git a/Justfile b/Justfile
index be484f8..59eb317 100644
--- a/Justfile
+++ b/Justfile
@@ -1,6 +1,10 @@
 set windows-shell := ["pwsh", "-c"]
 set dotenv-load := true
 export PYTHONPATH := "src"
+export PYTHONUTF8 := "1"
+export MLFLOW_TRACKING_URI := env_var("MLFLOW_TRACKING_URI")
+export MLFLOW_TRACKING_USERNAME := env_var("MLFLOW_TRACKING_USERNAME")
+export MLFLOW_TRACKING_PASSWORD := env_var("MLFLOW_TRACKING_PASSWORD")
 
 # Ensures Docker Desktop is running before any build step.
 # Attempts auto-launch on Windows; waits up to 90s for daemon to become ready.
@@ -40,7 +44,7 @@ train-prod:
 export-model:
 	uv run python -m scripts.export_model
 
-build-api: check-docker export-model
+build-api: check-docker
 	docker build -f Dockerfile.api -t rossmann-api:latest .
 
 build-ui: check-docker
diff --git a/scripts/clean_artifacts.py b/scripts/clean_artifacts.py
index 5de9eb6..9cf91e0 100644
--- a/scripts/clean_artifacts.py
+++ b/scripts/clean_artifacts.py
@@ -3,12 +3,25 @@
 from pathlib import Path
 
 
-def main():
+def _clean_dir(target: Path) -> None:
+    """Wipe non-.dvc files from a directory and remove empty subdirs."""
+    print(f"Cleaning directory: {target.name}/ (preserving .dvc files)")
+    for path in target.rglob("*"):
+        if path.is_file() and path.suffix != ".dvc":
+            try:
+                path.unlink()
+            except Exception as e:
+                print(f"  Failed delete {path}: {e}")
+    for path in sorted(target.rglob("*"), key=lambda p: len(p.parts), reverse=True):
+        if path.is_dir() and not any(path.iterdir()):
+            path.rmdir()
+
+
+def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument("--yes", action="store_true", help="Skip confirmation prompt")
     args = parser.parse_args()
 
-    # Dynamically resolve project root regardless of where script is called
     project_root = Path(__file__).resolve().parents[1]
 
     if not args.yes:
@@ -19,33 +32,15 @@ def main():
             print("Aborted.")
             return
 
-    dirs_to_process = ["mlruns", "mlartifacts", "models"]
-    files_to_delete = ["mlflow.db"] # Specifically target this outside if needed
-
-    for d in dirs_to_process:
+    for d in ["mlruns", "mlartifacts", "models"]:
         target = project_root / d
         if target.exists() and target.is_dir():
-            print(f"Cleaning directory: {d}/ (preserving .dvc files)")
-            # Recursive cleanup
-            for path in target.rglob("*"):
-                if path.is_file() and path.suffix != ".dvc":
-                    try:
-                        path.unlink()
-                        # print(f"  Deleted: {path.relative_to(project_root)}")
-                    except Exception as e:
-                        print(f"  Failed delete {path}: {e}")
-            
-            # Clean up empty subdirectories (except DVC metadata folders if any)
-            # We walk bottom-up to remove childless dirs
-            for path in sorted(target.rglob("*"), key=lambda p: len(p.parts), reverse=True):
-                if path.is_dir() and not any(path.iterdir()):
-                    path.rmdir()
-
-    for f in files_to_delete:
-        target = project_root / f
-        if target.exists() and target.is_file():
-            os.remove(target)
-            print(f"Deleted root-level file: {f}")
+            _clean_dir(target)
+
+    db = project_root / "mlflow.db"
+    if db.exists():
+        os.remove(db)
+        print("Deleted root-level file: mlflow.db")
 
     print("Wiped artifacts successfully.")
 
diff --git a/scripts/export_model.py b/scripts/export_model.py
index d8c8f88..1a4635b 100644
--- a/scripts/export_model.py
+++ b/scripts/export_model.py
@@ -13,17 +13,21 @@
 logger = logging.getLogger(__name__)
 
 
-def export_latest_model():
+def export_latest_model() -> None:
     """
-    LOCAL UTILITY — not called by the CI/CD pipeline.
+    LOCAL UTILITY — not called by the CI/CD pipeline or 'just deploy-all'.
 
     Finds the latest run in 'Rossmann_Production' and copies the model
     artifact from the MLflow artifact store to models/production_model/.
     Use this locally to swap in a specific historical run without retraining:
 
-        MLFLOW_TRACKING_URI=... uv run python scripts/export_model.py
+        uv run python -m scripts.export_model
 
-    In CI, train_model.py saves the model directly to models/ during training.
+    NOTE: DagsHub runs an MLflow 2.x server which does not support model
+    artifact storage from MLflow 3.x clients (artifact_path upload silently
+    fails). Model artifacts are versioned via DVC instead. Run 'just pull'
+    to restore the latest DVC-tracked model, or 'just train-prod' to train
+    a fresh one locally.
     """
     mlflow.set_tracking_uri(
         os.getenv("MLFLOW_TRACKING_URI", "sqlite:///mlruns/mlflow.db")
diff --git a/src/rossmann_ops/train_model.py b/src/rossmann_ops/train_model.py
index c6a10d1..e309c50 100644
--- a/src/rossmann_ops/train_model.py
+++ b/src/rossmann_ops/train_model.py
@@ -209,7 +209,7 @@ def train_production_model() -> None:
             mlflow.log_artifact(shap_tmp)
 
         # 13. Log Model to MLflow Registry
-        mlflow.sklearn.log_model(model, name="production_model")
+        mlflow.sklearn.log_model(model, artifact_path="production_model")
 
         # 14. Save Model Locally for Docker/CI builds
         local_model_dir = project_root / config["model"]["save_path"]