Update health check and cloud deployment

.env.example

@@ -71,6 +71,16 @@ EVALUATOR_LLM_MODEL=llama-3.3-70b-versatile
 # GROQ_MODEL=qwen/qwen3-32b
 # LOCAL_EMBEDDING_MODEL=paraphrase-multilingual-MiniLM-L12-v2
 
+# *****************************************************************************
+# EXAMPLE 2d — CLOUD MODE (AWS Bedrock)
+# *****************************************************************************
+# LLM_PROVIDER=bedrock
+# EMBEDDING_PROVIDER=bedrock
+# AWS_REGION=eu-west-1
+# AWS_BEDROCK_MODEL=anthropic.claude-sonnet-4-20250514-v1:0
+# AWS_BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+# Note: Uses default AWS credential chain (env vars, ~/.aws/credentials, or IAM role)
+
 # *****************************************************************************
 # EXAMPLE 2e — CLOUD MODE (Anthropic LLM + local embeddings)
 # *****************************************************************************

.github/README.md

@@ -15,8 +15,8 @@ A RAG application that lets users ask questions about documents in any language
 | Agent flows | Plan-and-Execute with six tools, ReAct sub-agent and conversation memory |
 | Evaluation | RAGAS metrics (faithfulness, answer relevancy, context precision) |
 | Traceability | Each answer includes source references with chunk ID and page number, plus structured logging |
-| Provider abstraction | Factory pattern that allows swapping between Ollama, OpenAI, Azure OpenAI, Anthropic and Google GenAI without touching business code |
-| Deployment | Docker Compose for local setup, Hugging Face Spaces for the public demo |
+| Provider abstraction | Factory pattern that allows swapping between Ollama, OpenAI, Azure OpenAI, AWS Bedrock, Anthropic and Google GenAI without touching business code |
+| Deployment | Docker Compose (local), Azure Container Apps, AWS ECS Fargate, Hugging Face Spaces (demo) |
 
 ### How it works
 
@@ -80,10 +80,29 @@ The test set is auto-generated and biased toward single-document factual questio
 
 ### Provider support
 
-LLM and embedding backends are configured through environment variables. Supported providers are Ollama, OpenAI, Azure OpenAI, Anthropic, Google GenAI and Groq. The default setup (Ollama and HuggingFace) runs entirely locally without any API keys.
+LLM and embedding backends are configured through environment variables. Supported providers are Ollama, OpenAI, Azure OpenAI, AWS Bedrock, Anthropic, Google GenAI and Groq. The default setup (Ollama and HuggingFace) runs entirely locally without any API keys.
 
 See `.env.example` for per-provider configuration.
 
+### Cloud deployment
+
+The application is cloud-agnostic by design. Business code depends only on LangChain abstract interfaces; the concrete provider is selected at deploy time via environment variables.
+
+| Layer | Azure | AWS | Local |
+|---|---|---|---|
+| LLM / Embeddings | Azure OpenAI | Bedrock (Claude, Titan) | Ollama + HuggingFace |
+| Container registry | ACR | ECR | - |
+| Runtime | Container Apps | ECS Fargate | docker-compose |
+| CI/CD | GitHub Actions | GitHub Actions | - |
+
+GitHub Actions workflows are included for both clouds:
+
+- `ci.yml` runs lint, type check, and tests on every push and PR
+- `deploy-azure.yml` builds, pushes to ACR, and deploys to Azure Container Apps
+- `deploy-aws.yml` builds, pushes to ECR, and deploys to ECS Fargate
+
+Health probes (`/health/live` for liveness, `/health/ready` for readiness) are used by container orchestrators to manage rolling deployments.
+
 ### Try it live
 
 The demo lives at [xq-dokumentassistent.hf.space](https://xq-dokumentassistent.hf.space).
@@ -164,20 +183,27 @@ src/
     reranker.py            # cross-encoder
   api/
     main.py
-    routes.py              # /query, /ingest, /health
+    routes.py              # /query, /ingest, /health/live, /health/ready
   agent/
     intent_classifier.py
     router.py              # pipeline mode (AGENT_MODE=pipeline)
     tools.py               # six retrieval tools and ToolResultStore
     plan_and_execute.py    # Plan-and-Execute agent (AGENT_MODE=react)
     memory.py              # conversation memory for multi-turn
+    session_store.py       # SQLite-backed per-session memory persistence
   evaluation/
     evaluator.py           # RAGAS metrics
   ui/
     app.py                 # Streamlit frontend
 scripts/
   ingest.py
+  evaluate.py              # RAGAS evaluation CLI
   e2e_test.py
 tests/
 docs/                      # example PDFs or texts (KU AI public documents)
+.github/
+  workflows/
+    ci.yml                 # lint + test on push/PR
+    deploy-azure.yml       # build, push ACR, deploy Container Apps
+    deploy-aws.yml         # build, push ECR, deploy ECS Fargate
 ```

.github/workflows/ci.yml

@@ -0,0 +1,49 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint-and-test:
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: pip
+
+      - name: Install dependencies
+        run: pip install -r requirements.txt
+
+      - name: Lint with ruff
+        run: |
+          pip install ruff
+          ruff check src/ tests/
+
+      - name: Type check with mypy
+        run: |
+          pip install mypy
+          mypy src/ --ignore-missing-imports --no-error-summary || true
+
+      - name: Run tests
+        run: pytest tests/ -v --ignore=tests/evaluation/
+        env:
+          LLM_PROVIDER: ollama
+          EMBEDDING_PROVIDER: local
+
+  build-image:
+    runs-on: ubuntu-latest
+    needs: lint-and-test
+    if: github.ref == 'refs/heads/main'
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Build Docker image
+        run: docker build -f Dockerfile.compose -t doc-assistant:${{ github.sha }} .

.github/workflows/deploy-aws.yml

@@ -0,0 +1,71 @@
+name: Deploy to AWS
+
+on:
+  # Uncomment to auto-deploy after CI passes on main:
+  # workflow_run:
+  #   workflows: [CI]
+  #   types: [completed]
+  #   branches: [main]
+  workflow_dispatch:
+
+env:
+  AWS_REGION: ${{ vars.AWS_REGION || 'eu-west-1' }}
+  ECR_REPOSITORY: doc-assistant
+  ECS_CLUSTER: ${{ vars.ECS_CLUSTER }}
+  ECS_SERVICE: doc-assistant
+  TASK_DEFINITION_FAMILY: doc-assistant
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    if: ${{ github.event.workflow_run.conclusion == 'success' || github.event_name == 'workflow_dispatch' }}
+
+    permissions:
+      id-token: write
+      contents: read
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          role-to-assume: ${{ secrets.AWS_ROLE_ARN }}
+          aws-region: ${{ env.AWS_REGION }}
+
+      - name: Login to ECR
+        id: ecr-login
+        uses: aws-actions/amazon-ecr-login@v2
+
+      - name: Build and push image to ECR
+        id: build
+        env:
+          ECR_REGISTRY: ${{ steps.ecr-login.outputs.registry }}
+        run: |
+          IMAGE_TAG=${{ env.ECR_REGISTRY }}/${{ env.ECR_REPOSITORY }}:${{ github.sha }}
+          docker build -f Dockerfile.compose -t $IMAGE_TAG .
+          docker push $IMAGE_TAG
+          echo "image=$IMAGE_TAG" >> $GITHUB_OUTPUT
+
+      - name: Download current task definition
+        run: |
+          aws ecs describe-task-definition \
+            --task-definition ${{ env.TASK_DEFINITION_FAMILY }} \
+            --query taskDefinition \
+            > task-definition.json
+
+      - name: Update task definition with new image
+        id: task-def
+        uses: aws-actions/amazon-ecs-render-task-definition@v1
+        with:
+          task-definition: task-definition.json
+          container-name: api
+          image: ${{ steps.build.outputs.image }}
+
+      - name: Deploy to ECS
+        uses: aws-actions/amazon-ecs-deploy-task-definition@v2
+        with:
+          task-definition: ${{ steps.task-def.outputs.task-definition }}
+          service: ${{ env.ECS_SERVICE }}
+          cluster: ${{ env.ECS_CLUSTER }}
+          wait-for-service-stability: true

.github/workflows/deploy-azure.yml

@@ -0,0 +1,61 @@
+name: Deploy to Azure
+
+on:
+  # Uncomment to auto-deploy after CI passes on main:
+  # workflow_run:
+  #   workflows: [CI]
+  #   types: [completed]
+  #   branches: [main]
+  workflow_dispatch:
+
+env:
+  AZURE_RESOURCE_GROUP: ${{ vars.AZURE_RESOURCE_GROUP }}
+  ACR_NAME: ${{ vars.ACR_NAME }}
+  CONTAINER_APP_NAME: doc-assistant
+  IMAGE_NAME: doc-assistant
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    if: ${{ github.event.workflow_run.conclusion == 'success' || github.event_name == 'workflow_dispatch' }}
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Login to Azure
+        uses: azure/login@v2
+        with:
+          creds: ${{ secrets.AZURE_CREDENTIALS }}
+
+      - name: Login to ACR
+        run: az acr login --name ${{ env.ACR_NAME }}
+
+      - name: Build and push image to ACR
+        run: |
+          IMAGE_TAG=${{ env.ACR_NAME }}.azurecr.io/${{ env.IMAGE_NAME }}:${{ github.sha }}
+          docker build -f Dockerfile.compose -t $IMAGE_TAG .
+          docker push $IMAGE_TAG
+
+      - name: Deploy to Container Apps
+        run: |
+          az containerapp update \
+            --name ${{ env.CONTAINER_APP_NAME }} \
+            --resource-group ${{ env.AZURE_RESOURCE_GROUP }} \
+            --image ${{ env.ACR_NAME }}.azurecr.io/${{ env.IMAGE_NAME }}:${{ github.sha }}
+
+      - name: Wait for readiness
+        run: |
+          APP_URL=$(az containerapp show \
+            --name ${{ env.CONTAINER_APP_NAME }} \
+            --resource-group ${{ env.AZURE_RESOURCE_GROUP }} \
+            --query properties.configuration.ingress.fqdn -o tsv)
+          for i in $(seq 1 30); do
+            if curl -sf "https://${APP_URL}/health/ready" > /dev/null 2>&1; then
+              echo "Application is ready"
+              exit 0
+            fi
+            echo "Waiting for readiness... ($i/30)"
+            sleep 10
+          done
+          echo "Readiness check timed out"
+          exit 1

README.md

@@ -25,8 +25,8 @@ A RAG application that lets users ask questions about documents in any language
 | Agent flows | Plan-and-Execute with six tools, ReAct sub-agent and conversation memory |
 | Evaluation | RAGAS metrics (faithfulness, answer relevancy, context precision) |
 | Traceability | Each answer includes source references with chunk ID and page number, plus structured logging |
-| Provider abstraction | Factory pattern that allows swapping between Ollama, OpenAI, Azure OpenAI, Anthropic and Google GenAI without touching business code |
-| Deployment | Docker Compose for local setup, Hugging Face Spaces for the public demo |
+| Provider abstraction | Factory pattern that allows swapping between Ollama, OpenAI, Azure OpenAI, AWS Bedrock, Anthropic and Google GenAI without touching business code |
+| Deployment | Docker Compose (local), Azure Container Apps, AWS ECS Fargate, Hugging Face Spaces (demo) |
 
 ### How it works
 
@@ -74,10 +74,29 @@ Configuration lives in environment variables via `src/config.py`; there are no h
 
 ### Provider support
 
-LLM and embedding backends are configured through environment variables. Supported providers are Ollama, OpenAI, Azure OpenAI, Anthropic, Google GenAI and Groq. The default setup (Ollama and HuggingFace) runs entirely locally without any API keys.
+LLM and embedding backends are configured through environment variables. Supported providers are Ollama, OpenAI, Azure OpenAI, AWS Bedrock, Anthropic, Google GenAI and Groq. The default setup (Ollama and HuggingFace) runs entirely locally without any API keys.
 
 See `.env.example` for per-provider configuration.
 
+### Cloud deployment
+
+The application is cloud-agnostic by design. Business code depends only on LangChain abstract interfaces; the concrete provider is selected at deploy time via environment variables.
+
+| Layer | Azure | AWS | Local |
+|---|---|---|---|
+| LLM / Embeddings | Azure OpenAI | Bedrock (Claude, Titan) | Ollama + HuggingFace |
+| Container registry | ACR | ECR | - |
+| Runtime | Container Apps | ECS Fargate | docker-compose |
+| CI/CD | GitHub Actions | GitHub Actions | - |
+
+GitHub Actions workflows are included for both clouds:
+
+- `ci.yml` runs lint, type check, and tests on every push and PR
+- `deploy-azure.yml` builds, pushes to ACR, and deploys to Azure Container Apps
+- `deploy-aws.yml` builds, pushes to ECR, and deploys to ECS Fargate
+
+Health probes (`/health/live` for liveness, `/health/ready` for readiness) are used by both Kubernetes and container orchestrators to manage rolling deployments.
+
 ### Try it live
 
 The demo lives at [xq-dokumentassistent.hf.space](https://xq-dokumentassistent.hf.space).
@@ -134,6 +153,56 @@ cp .env.example .env
 docker compose up --build
 ```
 
+#### Azure Container Apps
+
+```bash
+# Login and set variables
+az login
+ACR_NAME=<your-acr-name>
+RG=<your-resource-group>
+APP_NAME=doc-assistant
+
+# Build and push
+az acr login --name $ACR_NAME
+docker build -f Dockerfile.compose -t $ACR_NAME.azurecr.io/doc-assistant:latest .
+docker push $ACR_NAME.azurecr.io/doc-assistant:latest
+
+# Deploy
+az containerapp create \
+  --name $APP_NAME \
+  --resource-group $RG \
+  --image $ACR_NAME.azurecr.io/doc-assistant:latest \
+  --target-port 8000 \
+  --env-vars LLM_PROVIDER=azure_openai \
+    AZURE_OPENAI_API_KEY=secretref:aoai-key \
+    AZURE_OPENAI_ENDPOINT=https://<resource>.openai.azure.com/ \
+    AZURE_OPENAI_DEPLOYMENT=<deployment> \
+    AZURE_OPENAI_EMBEDDING_DEPLOYMENT=<embedding-deployment>
+```
+
+CI/CD is automated via `.github/workflows/deploy-azure.yml`.
+
+#### AWS ECS Fargate
+
+```bash
+# Login and set variables
+AWS_REGION=eu-west-1
+ECR_REPO=<account-id>.dkr.ecr.$AWS_REGION.amazonaws.com/doc-assistant
+
+# Build and push
+aws ecr get-login-password --region $AWS_REGION | docker login --username AWS --password-stdin $ECR_REPO
+docker build -f Dockerfile.compose -t $ECR_REPO:latest .
+docker push $ECR_REPO:latest
+
+# Deploy (update existing ECS service)
+aws ecs update-service \
+  --cluster doc-assistant \
+  --service doc-assistant \
+  --force-new-deployment
+```
+
+CI/CD is automated via `.github/workflows/deploy-aws.yml`. The task definition should set `LLM_PROVIDER=bedrock` and grant the task role Bedrock access.
+
 #### Hugging Face Spaces
 
 A `Dockerfile` and supervisor configuration are included. The Space runs Qdrant, the API and the UI behind nginx on port 7860.
@@ -158,21 +227,28 @@ src/
     reranker.py            # cross-encoder
   api/
     main.py
-    routes.py              # /query, /ingest, /health
+    routes.py              # /query, /ingest, /health/live, /health/ready
   agent/
     intent_classifier.py
     router.py              # pipeline mode (AGENT_MODE=pipeline)
     tools.py               # six retrieval tools and ToolResultStore
     plan_and_execute.py    # Plan-and-Execute agent (AGENT_MODE=react)
     memory.py              # conversation memory for multi-turn
+    session_store.py       # SQLite-backed per-session memory persistence
   evaluation/
     evaluator.py           # RAGAS metrics
   ui/
     app.py                 # Streamlit frontend
 scripts/
   ingest.py
+  evaluate.py              # RAGAS evaluation CLI
   e2e_test.py
 tests/
 docs/                      # example PDFs or texts (KU AI public documents)
+.github/
+  workflows/
+    ci.yml                 # lint + test on push/PR
+    deploy-azure.yml       # build, push ACR, deploy Container Apps
+    deploy-aws.yml         # build, push ECR, deploy ECS Fargate
 ```
 

docker-compose.yml

@@ -6,6 +6,11 @@ services:
       - "6334:6334"
     volumes:
       - qdrant_data:/qdrant/storage
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:6333/healthz"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
 
   ollama:
     image: ollama/ollama:latest
@@ -45,9 +50,16 @@ services:
       - QDRANT_URL=http://qdrant:6333
       - OLLAMA_BASE_URL=http://ollama:11434
     depends_on:
-      - qdrant
+      qdrant:
+        condition: service_healthy
     volumes:
       - ./docs:/app/docs
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:8000/health/ready"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 60s
 
   ui:
     build:
@@ -60,7 +72,8 @@ services:
     environment:
       - API_BASE_URL=http://api:8000
     depends_on:
-      - api
+      api:
+        condition: service_healthy
     entrypoint: []
     command: ["streamlit", "run", "src/ui/app.py", "--server.port=8501", "--server.address=0.0.0.0"]
 

requirements.txt

@@ -11,6 +11,7 @@ langchain-ollama==1.0.1
 langchain-huggingface==1.2.1
 langchain-anthropic==1.4.0
 langchain-google-genai==4.2.1
+langchain-aws==0.2.17
 langchain-text-splitters==1.1.1
 langchain-experimental==0.4.1
 

src/api/routes.py

@@ -172,13 +172,15 @@ class HealthResponse(BaseModel):
     embedding_model: str = ""
 
 
-@router.get("/health", response_model=HealthResponse)
-async def health_check() -> HealthResponse:
-    """Health check endpoint.
+class ReadinessResponse(BaseModel):
+    """Response body for the readiness probe."""
 
-    Returns:
-        HealthResponse with service status and version.
-    """
+    status: str
+    checks: dict[str, bool]
+
+
+def _build_health_response() -> HealthResponse:
+    """Build the full health response with provider details."""
     llm_provider = ""
     llm_model = ""
     embedding_provider = ""
@@ -191,6 +193,7 @@ async def health_check() -> HealthResponse:
             "ollama": _settings.ollama_model,
             "openai": _settings.openai_model,
             "azure_openai": _settings.azure_openai_deployment,
+            "bedrock": _settings.aws_bedrock_model,
             "groq": _settings.groq_model,
             "anthropic": _settings.anthropic_model,
             "google_genai": _settings.google_model,
@@ -206,6 +209,67 @@ async def health_check() -> HealthResponse:
     )
 
 
+@router.get("/health", response_model=HealthResponse)
+async def health_check() -> HealthResponse:
+    """Health check endpoint (backwards compatible).
+
+    Returns:
+        HealthResponse with service status and version.
+    """
+    return _build_health_response()
+
+
+@router.get("/health/live", response_model=HealthResponse)
+async def liveness() -> HealthResponse:
+    """Liveness probe. Returns 200 if the process is running.
+
+    Kubernetes uses this to decide whether to restart the container.
+    Does not check external dependencies.
+
+    Returns:
+        HealthResponse with service status and version.
+    """
+    return _build_health_response()
+
+
+@router.get("/health/ready", response_model=ReadinessResponse)
+async def readiness() -> ReadinessResponse:
+    """Readiness probe. Returns 200 only when all dependencies are available.
+
+    Kubernetes uses this to decide whether to route traffic to the pod.
+    Checks: vector store reachable, BM25 index loaded.
+
+    Returns:
+        ReadinessResponse with per-dependency check results.
+
+    Raises:
+        HTTPException: 503 if any dependency check fails.
+    """
+    checks: dict[str, bool] = {}
+
+    # Check vector store connectivity
+    try:
+        if _vector_store is not None:
+            _vector_store.get_all_chunks()[:0]  # lightweight probe
+            checks["vector_store"] = True
+        else:
+            checks["vector_store"] = False
+    except Exception:
+        logger.warning("Readiness check failed: vector store unreachable")
+        checks["vector_store"] = False
+
+    # Check BM25 index is loaded
+    checks["bm25_index"] = _bm25_search is not None and _bm25_search.is_indexed
+
+    # Check router is wired up
+    checks["router"] = _query_router is not None
+
+    all_ready = all(checks.values())
+    if not all_ready:
+        raise HTTPException(status_code=503, detail={"status": "unavailable", "checks": checks})
+    return ReadinessResponse(status="ready", checks=checks)
+
+
 @router.post("/query", response_model=QueryResponse)
 async def query_documents(request: QueryRequest) -> QueryResponse:
     """Query the document knowledge base.

src/config.py

@@ -57,6 +57,11 @@ class Settings:
     azure_openai_deployment: str
     azure_openai_embedding_deployment: str
 
+    # AWS Bedrock
+    aws_region: str
+    aws_bedrock_model: str
+    aws_bedrock_embedding_model: str
+
     # Groq
     groq_api_key: str
     groq_model: str
@@ -143,6 +148,11 @@ def load_settings() -> Settings:
         azure_openai_deployment=os.environ.get("AZURE_OPENAI_DEPLOYMENT", ""),
         azure_openai_embedding_deployment=os.environ.get("AZURE_OPENAI_EMBEDDING_DEPLOYMENT", ""),
 
+        # AWS Bedrock
+        aws_region=os.environ.get("AWS_REGION", "eu-west-1"),
+        aws_bedrock_model=os.environ.get("AWS_BEDROCK_MODEL", "anthropic.claude-sonnet-4-20250514-v1:0"),
+        aws_bedrock_embedding_model=os.environ.get("AWS_BEDROCK_EMBEDDING_MODEL", "amazon.titan-embed-text-v2:0"),
+
         # Groq
         groq_api_key=os.environ.get("GROQ_API_KEY", ""),
         groq_model=os.environ.get("GROQ_MODEL", "qwen/qwen3-32b"),

src/provider.py

@@ -14,8 +14,8 @@ from src.config import Settings
 
 logger = logging.getLogger(__name__)
 
-_SUPPORTED_LLM_PROVIDERS = ["ollama", "azure_openai", "openai", "groq", "anthropic", "google_genai"]
-_SUPPORTED_EMBEDDING_PROVIDERS = ["local", "azure_openai", "openai", "google_genai"]
+_SUPPORTED_LLM_PROVIDERS = ["ollama", "azure_openai", "openai", "groq", "anthropic", "google_genai", "bedrock"]
+_SUPPORTED_EMBEDDING_PROVIDERS = ["local", "azure_openai", "openai", "google_genai", "bedrock"]
 
 
 def create_llm(settings: Settings) -> BaseChatModel:
@@ -94,6 +94,15 @@ def create_llm(settings: Settings) -> BaseChatModel:
                 temperature=0.0,
             )
 
+        case "bedrock":
+            from langchain_aws import ChatBedrockConverse
+
+            return ChatBedrockConverse(
+                model=settings.aws_bedrock_model,
+                region_name=settings.aws_region,
+                temperature=0.0,
+            )
+
         case _:
             raise ValueError(
                 f"Unknown LLM provider: '{provider}'. "
@@ -107,6 +116,7 @@ _EVALUATOR_MODEL_FIELD: dict[str, str] = {
     "anthropic": "anthropic_model",
     "google_genai": "google_model",
     "azure_openai": "azure_openai_deployment",
+    "bedrock": "aws_bedrock_model",
     "ollama": "ollama_model",
 }
 
@@ -200,6 +210,14 @@ def create_embeddings(settings: Settings) -> Embeddings:
                 google_api_key=settings.google_api_key,
             )
 
+        case "bedrock":
+            from langchain_aws import BedrockEmbeddings
+
+            return BedrockEmbeddings(
+                model_id=settings.aws_bedrock_embedding_model,
+                region_name=settings.aws_region,
+            )
+
         case _:
             raise ValueError(
                 f"Unknown embedding provider: '{provider}'. "

src/retrieval/bm25_search.py

@@ -17,6 +17,11 @@ class BM25Search:
         self._chunks: list[DocumentChunk] = []
         self._index: BM25Okapi | None = None
 
+    @property
+    def is_indexed(self) -> bool:
+        """Return True if the BM25 index has been built."""
+        return self._index is not None and len(self._chunks) > 0
+
     def index(self, chunks: list[DocumentChunk]) -> None:
         """Build the BM25 index from document chunks.
 

tests/test_api.py

@@ -5,7 +5,7 @@ from unittest.mock import MagicMock, patch
 import pytest
 from fastapi.testclient import TestClient
 
-from src.api.routes import router, set_dependencies, QueryResponse, HealthResponse
+from src.api.routes import router, set_dependencies, QueryResponse, HealthResponse, ReadinessResponse
 from src.models import DocumentChunk, GenerationResponse, IntentType, QueryResult
 
 
@@ -65,6 +65,46 @@ class TestHealthCheck:
         assert body["version"] == "0.1.0"
 
 
+class TestLivenessProbe:
+    """Tests for the /health/live endpoint."""
+
+    def test_liveness_returns_ok(self, client: TestClient) -> None:
+        response = client.get("/health/live")
+
+        assert response.status_code == 200
+        body = response.json()
+        assert body["status"] == "ok"
+
+
+class TestReadinessProbe:
+    """Tests for the /health/ready endpoint."""
+
+    def test_readiness_returns_ready_when_all_deps_available(
+        self, client: TestClient, mock_deps: dict[str, MagicMock]
+    ) -> None:
+        mock_deps["vector_store"].get_all_chunks.return_value = []
+        mock_deps["bm25_search"].is_indexed = True
+
+        response = client.get("/health/ready")
+
+        assert response.status_code == 200
+        body = response.json()
+        assert body["status"] == "ready"
+        assert body["checks"]["vector_store"] is True
+        assert body["checks"]["bm25_index"] is True
+        assert body["checks"]["router"] is True
+
+    def test_readiness_returns_503_when_bm25_not_indexed(
+        self, client: TestClient, mock_deps: dict[str, MagicMock]
+    ) -> None:
+        mock_deps["vector_store"].get_all_chunks.return_value = []
+        mock_deps["bm25_search"].is_indexed = False
+
+        response = client.get("/health/ready")
+
+        assert response.status_code == 503
+
+
 class TestQueryEndpoint:
     """Tests for the /query endpoint."""
 

tests/test_bm25.py

@@ -14,6 +14,15 @@ def _make_chunk(chunk_id: str, text: str) -> DocumentChunk:
 class TestBM25Index:
     """Tests for index construction."""
 
+    def test_is_indexed_false_before_indexing(self) -> None:
+        bm25 = BM25Search()
+        assert bm25.is_indexed is False
+
+    def test_is_indexed_true_after_indexing(self) -> None:
+        bm25 = BM25Search()
+        bm25.index([_make_chunk("1", "hello world")])
+        assert bm25.is_indexed is True
+
     def test_index_stores_chunks(self) -> None:
         bm25 = BM25Search()
         chunks = [_make_chunk("1", "hello world"), _make_chunk("2", "foo bar")]