fix(cookbook): address reviewer feedback on rag_gateway example

- Add @component decorator and @component.output_types to LiteLLMGenerator
  for Haystack 2.0 compatibility (was causing runtime crash)
- Remove unused InMemoryBM25Retriever import from rag_haystack.py
- Pin ragas>=0.1.0,<0.2.0 to avoid API incompatibility with 0.2.x+
- Wire eval_model parameter through to RAGAS via LangchainLLMWrapper
- Remove silent dummy API key fallback; raise explicit ValueError instead
- Rename litellm_config.yaml to litellm_config.example.yaml (was gitignored)
- Update README references accordingly

cookbook/rag_gateway/README.md

@@ -19,7 +19,7 @@ This cookbook showcases:
 rag_gateway/
 ├── README.md                      # This file
 ├── requirements.txt               # Python dependencies
-├── litellm_config.yaml           # Multi-provider configuration (optional)
+├── litellm_config.example.yaml   # Multi-provider configuration (optional)
 ├── rag_llamaindex.py             # LlamaIndex RAG pipeline
 ├── rag_haystack.py               # Haystack RAG pipeline
 ├── evaluate_with_ragas.py        # RAGAS evaluation script
@@ -166,7 +166,7 @@ llm = setup_litellm(model_name="azure/gpt-4-deployment-name")
 
 ### Using Configuration File (Optional)
 
-The `litellm_config.yaml` file demonstrates advanced configuration:
+The `litellm_config.example.yaml` file demonstrates advanced configuration:
 
 ```yaml
 model_list:

cookbook/rag_gateway/evaluate_with_ragas.py

@@ -172,11 +172,22 @@ def evaluate_rag_system(
     """
     print(f"\n📊 Evaluating with RAGAS (using {eval_model} as judge)...")
     
+    # Require API key explicitly — do not silently fall back to a dummy value
+    if not os.getenv("OPENAI_API_KEY"):
+        raise ValueError(
+            "OPENAI_API_KEY is not set. "
+            "RAGAS requires an LLM to act as a judge. "
+            "Set it with: export OPENAI_API_KEY='your-key-here'"
+        )
+    
     # Convert to RAGAS dataset format
     dataset = Dataset.from_dict(rag_data)
     
-    # Configure RAGAS to use LiteLLM
-    os.environ["OPENAI_API_KEY"] = os.getenv("OPENAI_API_KEY", "dummy-key")
+    # Wire eval_model into RAGAS via LangchainLLMWrapper so the user-specified
+    # judge model is actually used instead of the RAGAS default.
+    from langchain_openai import ChatOpenAI
+    from ragas.llms import LangchainLLMWrapper
+    eval_llm = LangchainLLMWrapper(ChatOpenAI(model=eval_model))
     
     # Run evaluation
     try:
@@ -188,6 +199,7 @@ def evaluate_rag_system(
                 context_precision,
                 context_recall,
             ],
+            llm=eval_llm,
         )
         
         scores = {

cookbook/rag_gateway/litellm_config.example.yaml

@@ -0,0 +1,94 @@
+# LiteLLM Multi-Provider Configuration
+# This config demonstrates production-ready patterns for RAG systems
+
+model_list:
+  # Primary: OpenAI GPT-4 (high quality, higher cost)
+  - model_name: gpt-4-turbo
+    litellm_params:
+      model: gpt-4-turbo-preview
+      api_key: os.environ/OPENAI_API_KEY
+      timeout: 30  # 30 second timeout
+      max_retries: 3  # Retry up to 3 times
+      
+  # Fallback 1: OpenAI GPT-3.5 (faster, cheaper)
+  - model_name: gpt-3.5-turbo
+    litellm_params:
+      model: gpt-3.5-turbo
+      api_key: os.environ/OPENAI_API_KEY
+      timeout: 20
+      max_retries: 2
+      
+  # Fallback 2: Anthropic Claude (alternative provider)
+  - model_name: claude-3-haiku
+    litellm_params:
+      model: claude-3-haiku-20240307
+      api_key: os.environ/ANTHROPIC_API_KEY
+      timeout: 30
+      max_retries: 2
+      
+  # Local option: Ollama (free, private, no API key needed)
+  - model_name: ollama-llama3
+    litellm_params:
+      model: ollama/llama3
+      api_base: http://localhost:11434
+      timeout: 60  # Local models may be slower
+      
+  # Cost-effective option: Groq (very fast inference)
+  - model_name: groq-llama3
+    litellm_params:
+      model: groq/llama3-70b-8192
+      api_key: os.environ/GROQ_API_KEY
+      timeout: 15  # Groq is very fast
+      max_retries: 2
+
+# Router configuration for load balancing and fallbacks
+router_settings:
+  # Routing strategy: 'simple-shuffle', 'least-busy', 'usage-based-routing'
+  routing_strategy: simple-shuffle
+  
+  # Retry configuration
+  num_retries: 3
+  retry_after: 2  # Wait 2 seconds before retry
+  
+  # Timeout for entire request (including retries)
+  timeout: 45
+  
+  # Fallback models (in order of preference)
+  fallbacks:
+    - gpt-4-turbo
+    - gpt-3.5-turbo
+    - claude-3-haiku
+    - groq-llama3
+    - ollama-llama3
+
+# Logging configuration
+litellm_settings:
+  # Log level: DEBUG, INFO, WARNING, ERROR
+  set_verbose: true
+  
+  # Drop parameters not supported by provider
+  drop_params: true
+  
+  # Success callback for logging
+  success_callback: ["langfuse"]  # Optional: integrate with Langfuse for observability
+  
+  # Cost tracking
+  track_cost_per_request: true
+  
+  # Cache responses (optional, requires Redis)
+  # cache: true
+  # cache_params:
+  #   type: redis
+  #   host: localhost
+  #   port: 6379
+
+# General settings
+general_settings:
+  # Master key for proxy authentication (if using proxy server)
+  # master_key: os.environ/LITELLM_MASTER_KEY
+  
+  # Database for storing request logs (optional)
+  # database_url: os.environ/DATABASE_URL
+  
+  # Enable request/response logging
+  store_model_in_db: true

cookbook/rag_gateway/rag_haystack.py

@@ -13,8 +13,7 @@ import json
 from pathlib import Path
 from typing import List, Dict, Any
 
-from haystack import Pipeline, Document
-from haystack.components.retrievers.in_memory import InMemoryBM25Retriever
+from haystack import Pipeline, Document, component
 from haystack.components.builders import PromptBuilder
 from haystack.document_stores.in_memory import InMemoryDocumentStore
 from haystack.components.embedders import SentenceTransformersDocumentEmbedder, SentenceTransformersTextEmbedder
@@ -45,6 +44,7 @@ def load_documents(data_path: str = "sample_data/documents.json") -> List[Docume
     return documents
 
 
+@component
 class LiteLLMGenerator:
     """
     Custom Haystack component that uses LiteLLM for generation.
@@ -73,6 +73,7 @@ class LiteLLMGenerator:
         
         print(f"✅ Configured LiteLLM Generator with model: {model}")
     
+    @component.output_types(replies=List[str], meta=Dict[str, Any])
     def run(self, prompt: str) -> Dict[str, Any]:
         """
         Generate response using LiteLLM.

cookbook/rag_gateway/requirements.txt

@@ -9,7 +9,7 @@ llama-index-vector-stores-chroma>=0.1.0
 haystack-ai>=2.0.0
 
 # Evaluation
-ragas>=0.1.0
+ragas>=0.1.0,<0.2.0
 datasets>=2.14.0
 
 # Vector Store & Embeddings