Get multiple AI responses in parallel, then use another LLM to pick the best one using AWS Lambda durable functions
This pattern demonstrates the LLM-as-judge pattern with AWS Lambda durable functions. Multiple models answer a question in parallel, then another LLM evaluates and selects the best response.
All candidate answers and the final judgment are checkpointed for deterministic replay.
Benefits: Parallel execution reduces total workflow time. Checkpointing prevents re-running expensive LLM calls on retry. Deterministic replay ensures consistent results.
Select language:
Usage Example with Python:
LLM as judge with Lambda durable functions using Python.
import json
import re
import boto3
from aws_durable_execution_sdk_python import DurableContext, durable_execution
MODELS = ["us.amazon.nova-lite-v1:0", "us.amazon.nova-pro-v1:0"]
bedrock = boto3.client("bedrock-runtime")
def converse(model_id: str, prompt: str) -> str:
response = bedrock.converse(
modelId=model_id,
messages=[{"role": "user", "content": [{"text": prompt}]}],
)
return response["output"]["message"]["content"][0]["text"]
@durable_execution
def handler(event: dict, context: DurableContext):
question = event.get(
"question", "Write a 100 word summary on the great fire of London"
)
result = context.map(
MODELS,
lambda ctx, model_id, idx, items: {
"modelId": model_id,
"answer": converse(model_id, question),
},
"Get candidate answers",
)
candidates = result.get_results()
def judge(_):
responses = "\n".join(
f"{i + 1}. (Model: {c['modelId']}) {c['answer']}"
for i, c in enumerate(candidates)
)
prompt = f'''Question: "{question}"
Responses:
{responses}
Which response is best? Reply with JSON: {{"bestIndex": <1-based index>, "reasoning": "<why>"}}'''
response = converse(MODELS[0], prompt)
match = re.search(r"\{[\s\S]*\}", response)
parsed = json.loads(match.group(0)) if match else {}
best_index = (parsed.get("bestIndex", 1)) - 1
best = candidates[best_index] if 0 <= best_index < len(candidates) else candidates[0]
return {
"bestAnswer": best["answer"],
"reasoning": parsed.get("reasoning", ""),
"sourceModel": best["modelId"],
}
judgment = context.step(judge, "judge")
return {"question": question, **judgment}
Created by:
