feat(llm-katan): Add Kubernetes deployment support

e2e-tests/llm-katan/deploy/kubernetes/base/deployment.yaml

@@ -0,0 +1,144 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: llm-katan
+spec:
+  selector:
+    matchLabels: {}
+  replicas: 1
+  template:
+    metadata:
+      labels: {}
+    spec:
+      # Create a non-root user for security (matching Dockerfile)
+      securityContext:
+        fsGroup: 1000
+        runAsUser: 1000
+        runAsNonRoot: true
+
+      initContainers:
+        # Pre-download model to cache for faster startup
+        # Uses lightweight python:3.11-slim image and checks if model exists before downloading
+        - name: model-downloader
+          image: python:3.11-slim
+          imagePullPolicy: IfNotPresent
+          securityContext:
+            runAsUser: 0  # Run as root to install packages
+            runAsNonRoot: false
+            allowPrivilegeEscalation: false
+          command: ["/bin/bash", "-c"]
+          args:
+            - |
+              set -e
+
+              MODEL_ID="${YLLM_MODEL:-Qwen/Qwen3-0.6B}"
+              MODEL_DIR=$(basename "$MODEL_ID")
+
+              mkdir -p /cache/models
+              cd /cache/models
+
+              # Check if model already exists in PVC
+              if [ -d "$MODEL_DIR" ]; then
+                echo "Model $MODEL_ID already cached. Skipping download."
+                exit 0
+              fi
+
+              # Model not found, proceed with download
+              echo "Downloading model $MODEL_ID..."
+              pip install --no-cache-dir huggingface_hub[cli]
+              hf download "$MODEL_ID" --local-dir "$MODEL_DIR"
+          env:
+            - name: YLLM_MODEL
+              value: "Qwen/Qwen3-0.6B"
+            - name: HF_HUB_CACHE
+              value: "/tmp/hf_cache"
+          volumeMounts:
+            - name: models-volume
+              mountPath: /cache/models
+          resources:
+            requests:
+              memory: "512Mi"
+              cpu: "250m"
+            limits:
+              memory: "1Gi"
+              cpu: "500m"
+
+      containers:
+        - name: llm-katan
+          image: llm-katan:latest
+          imagePullPolicy: IfNotPresent
+
+          # Command is set via environment variables
+          # Default: llm-katan --model Qwen/Qwen3-0.6B --host 0.0.0.0 --port 8000
+
+          ports:
+            - name: http
+              containerPort: 8000
+              protocol: TCP
+
+          env:
+            # These can be overridden via ConfigMap in overlays
+            - name: YLLM_MODEL
+              value: "/cache/models/Qwen3-0.6B"  # Local path to downloaded model
+            - name: YLLM_PORT
+              value: "8000"
+            - name: YLLM_HOST
+              value: "0.0.0.0"
+            - name: YLLM_BACKEND
+              value: "transformers"
+            - name: PYTHONUNBUFFERED
+              value: "1"
+            - name: PYTHONDONTWRITEBYTECODE
+              value: "1"
+
+          volumeMounts:
+            - name: models-volume
+              mountPath: /cache/models
+
+          livenessProbe:
+            httpGet:
+              path: /health
+              port: http
+            initialDelaySeconds: 15
+            periodSeconds: 20
+            timeoutSeconds: 5
+            failureThreshold: 3
+
+          readinessProbe:
+            httpGet:
+              path: /health
+              port: http
+            initialDelaySeconds: 5
+            periodSeconds: 10
+            timeoutSeconds: 3
+            failureThreshold: 3
+
+          startupProbe:
+            httpGet:
+              path: /health
+              port: http
+            initialDelaySeconds: 30
+            periodSeconds: 15
+            timeoutSeconds: 5
+            failureThreshold: 60  # 15 minutes max startup time (for slow model downloads)
+
+          resources:
+            requests:
+              memory: "3Gi"
+              cpu: "1"
+            limits:
+              memory: "6Gi"
+              cpu: "2"
+
+          securityContext:
+            allowPrivilegeEscalation: false
+            readOnlyRootFilesystem: false  # HuggingFace needs to write to cache
+            runAsNonRoot: true
+            capabilities:
+              drop:
+                - ALL
+
+      volumes:
+        - name: models-volume
+          persistentVolumeClaim:
+            claimName: llm-katan-models

e2e-tests/llm-katan/deploy/kubernetes/base/kustomization.yaml

@@ -0,0 +1,21 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+
+metadata:
+  name: llm-katan-base
+
+namespace: llm-katan-system
+
+
+resources:
+  - namespace.yaml
+  - pvc.yaml
+  - deployment.yaml
+  - service.yaml
+
+# Images (can be overridden in overlays)
+images:
+  - name: llm-katan
+    newName: llm-katan
+    newTag: latest
+

e2e-tests/llm-katan/deploy/kubernetes/base/namespace.yaml

@@ -0,0 +1,4 @@
+apiVersion: v1
+kind: Namespace
+metadata:
+  name: llm-katan-system

e2e-tests/llm-katan/deploy/kubernetes/base/pvc.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: llm-katan-models
+spec:
+  accessModes:
+    - ReadWriteOnce
+  resources:
+    requests:
+      storage: 5Gi  # Increased for model cache (~600MB model + overhead)

e2e-tests/llm-katan/deploy/kubernetes/base/service.yaml

@@ -0,0 +1,14 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: llm-katan
+spec:
+  type: ClusterIP
+  selector:
+    app: llm-katan
+  ports:
+    - name: http
+      port: 8000
+      targetPort: http
+      protocol: TCP
+

e2e-tests/llm-katan/deploy/kubernetes/components/common/kustomization.yaml

@@ -0,0 +1,10 @@
+apiVersion: kustomize.config.k8s.io/v1alpha1
+kind: Component
+
+# Common labels applied to all resources that use this component
+labels:
+- includeSelectors: true
+  pairs:
+    app.kubernetes.io/name: llm-katan
+    app.kubernetes.io/part-of: semantic-router-workspaces
+    app.kubernetes.io/managed-by: kustomize

e2e-tests/llm-katan/deploy/kubernetes/overlays/claude/kustomization.yaml

@@ -0,0 +1,42 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+
+metadata:
+  name: llm-katan-claude
+
+resources:
+  - ../../base
+
+components:
+  - ../../components/common
+
+nameSuffix: -claude
+
+patches:
+  - target:
+      kind: Deployment
+      name: llm-katan
+    patch: |-
+      - op: add
+        path: /spec/template/spec/containers/0/env/-
+        value:
+          name: YLLM_SERVED_MODEL_NAME
+          value: "claude-3-haiku-20240307"
+      - op: add
+        path: /spec/template/metadata/labels/model-alias
+        value: "claude-3-haiku"
+  - target:
+      kind: Service
+      name: llm-katan
+    patch: |-
+      - op: add
+        path: /metadata/labels/model-alias
+        value: "claude-3-haiku"
+  # Update PVC reference in deployment to match suffixed PVC name
+  - target:
+      kind: Deployment
+      name: llm-katan
+    patch: |-
+      - op: replace
+        path: /spec/template/spec/volumes/0/persistentVolumeClaim/claimName
+        value: llm-katan-models-claude

e2e-tests/llm-katan/deploy/kubernetes/overlays/gpt35/kustomization.yaml

@@ -0,0 +1,41 @@
+apiVersion: kustomize.config.k8s.io/v1beta1
+kind: Kustomization
+
+resources:
+  - ../../base
+
+components:
+  - ../../components/common
+
+nameSuffix: -gpt35
+
+patches:
+  - target:
+      kind: Deployment
+      name: llm-katan
+    patch: |-
+      - op: add
+        path: /spec/template/spec/containers/0/env/-
+        value:
+          name: YLLM_SERVED_MODEL_NAME
+          value: "gpt-3.5-turbo"
+      - op: add
+        path: /spec/template/metadata/labels/model-alias
+        value: "gpt-3.5-turbo"
+
+  - target:
+      kind: Service
+      name: llm-katan
+    patch: |-
+      - op: add
+        path: /metadata/labels/model-alias
+        value: "gpt-3.5-turbo"
+
+  # Update PVC reference in deployment to match suffixed PVC name
+  - target:
+      kind: Deployment
+      name: llm-katan
+    patch: |-
+      - op: replace
+        path: /spec/template/spec/volumes/0/persistentVolumeClaim/claimName
+        value: llm-katan-models-gpt35

e2e-tests/llm-katan/llm_katan/config.py

@@ -31,15 +31,14 @@ def __post_init__(self):
 
         # Apply environment variable overrides
         self.model_name = os.getenv("YLLM_MODEL", self.model_name)
+        self.served_model_name = os.getenv("YLLM_SERVED_MODEL_NAME", self.served_model_name)
         self.port = int(os.getenv("YLLM_PORT", str(self.port)))
         self.backend = os.getenv("YLLM_BACKEND", self.backend)
         self.host = os.getenv("YLLM_HOST", self.host)
 
         # Validate backend
         if self.backend not in ["transformers", "vllm"]:
-            raise ValueError(
-                f"Invalid backend: {self.backend}. Must be 'transformers' or 'vllm'"
-            )
+            raise ValueError(f"Invalid backend: {self.backend}. Must be 'transformers' or 'vllm'")
 
     @property
     def device_auto(self) -> str: