Exposing Your Bedrock App Using API Gateway Part 2

​

How API Gateway routing works

https://abc1234.execute-api.us-east-1.amazonaws.com/preprod

GET https://abc1234.execute-api.us-east-1.amazonaws.com/preprod/summarize
GET https://abc1234.execute-api.us-east-1.amazonaws.com/preprod/sentiment

​

Example pattern: one Lambda per resource/method

• Summarize Lambda: constructs a prompt and calls a Bedrock foundation model to produce a concise summary.
• Sentiment Lambda: constructs a prompt and calls a Bedrock foundation model to return sentiment (positive / negative / neutral).

1. Client calls API Gateway endpoint.
2. API Gateway routes the request to the configured Lambda.
3. Lambda constructs a prompt, calls Bedrock, and returns the model response.
4. API Gateway returns the response to the client.

import json
import os
import boto3

bedrock = boto3.client("bedrock-runtime")  # service name may vary by SDK version

MODEL_ID = os.environ.get("MODEL_ID", "my-foundation-model")

def lambda_handler(event, context):
    # Extract input from query string or mapped body
    text = event.get("queryStringParameters", {}).get("text", "") or json.loads(event.get("body", "{}")).get("text", "")
    prompt = f"Summarize this text:\n\n{text}"

    response = bedrock.invoke_model(
        modelId=MODEL_ID,
        contentType="application/json",
        accept="application/json",
        body=json.dumps({"prompt": prompt, "maxTokens": 200})
    )

    model_output = json.loads(response["body"])  # adjust parsing to the model's response shape

    return {
        "statusCode": 200,
        "headers": {"Content-Type": "application/json"},
        "body": json.dumps({"summary": model_output.get("output", "")})
    }

​

Event-driven vs API-based on-demand

• Event-driven (asynchronous): Infrastructure events (for example, S3 object-created notifications) trigger a Lambda that reads the object, builds a prompt, calls Bedrock, and stores or forwards the result.
• API-based on-demand (synchronous): A user or client calls an API Gateway endpoint; API Gateway routes the request to Lambda, which calls Bedrock and returns the response to the client.

​

Benefits of exposing Bedrock via API Gateway + Lambda

• Faster integration into existing systems: Consumers call a single REST API to access generative features.
• Reuse: Front ends, mobile apps, and third-party services share the same backend logic.
• Lower maintenance overhead: Fixes and model tuning happen server-side and benefit all clients.
• Interface flexibility: Expose simple REST routes (summarize, sentiment, classify) while encapsulating model details.

​

Key implementation notes (practical tips)

• Use structured request and response schemas (JSON) so clients know the expected payload shape.
• Choose the right API Gateway integration:
  • Lambda proxy integration forwards the full HTTP request to Lambda (headers, path, query string, body) — implement parsing in Lambda.
  • Mapped integration lets API Gateway transform requests/responses before they reach the Lambda.
• Keep configuration and secrets out of client code. Use Lambda environment variables, AWS Secrets Manager, or a configuration service for model IDs and sensitive parameters.
• Monitor costs and latency. Foundation models can vary in performance and cost — track usage with Amazon CloudWatch and use AWS X-Ray to trace end-to-end latency.
• Implement retries and graceful error handling in Lambda to handle transient model invocation failures.
• Secure your API Gateway endpoints with authentication and authorization: API keys, IAM, Amazon Cognito, or custom authorizers.

​

Quick checklist before production

• Define clear input/output JSON contracts for each route.
• Store model identifiers and invocation parameters in a secure configuration store.
• Apply least-privilege IAM roles to Lambdas and API Gateway.
• Set up CloudWatch alarms and X-Ray tracing for observability.
• Add request throttling and quotas in API Gateway to avoid runaway costs.

​

Takeaway

​

Links and references

• Amazon API Gateway documentation
• AWS Lambda documentation
• Amazon Bedrock documentation
• Amazon S3 documentation
• AWS Secrets Manager
• Amazon CloudWatch
• AWS X-Ray
• Amazon Cognito