#2- Complete End To End Generative AI Project On AWS Using AWS Bedrock And AWS Lambda

The core build is an end-to-end “blog generator” API on AWS: Postman sends a blog topic to an API Gateway endpoint, an AWS Lambda function uses Amazon Bedrock to generate a 200-word blog with a foundation model (Llama 2 chat), and the resulting text is saved to an S3 bucket as a timestamped .txt file. The workflow ties together API Gateway → Lambda → Bedrock → S3, turning a single HTTP request into a stored generative output.

Implementation starts with Amazon Bedrock model access. In the Bedrock console, the process requires granting model access per region (the walkthrough uses us-east-1). Some models require additional use-case details before access is granted, so the setup includes selecting which foundation models are allowed and saving the changes. The foundation model catalog includes options such as Llama 2, Llama 3, Titan (Amazon), Claude (Anthropic), Command (Cohere), Jurassic (AI21 Labs), Mistral, and Stable Diffusion for image generation—though the blog generation flow later focuses on a Llama 2 chat model.

Next comes AWS Lambda, set up as the compute layer that reacts to API calls. The function is created with Python 3.12 and a timeout configured to 3 minutes. Because Bedrock invocation requires a newer boto3 than the default Lambda runtime provides, the walkthrough adds a custom Lambda Layer containing an updated boto3 package compatible with Python 3.12 (and also mentions Python 3.10/3.11 compatibility during layer creation). The Lambda code uses boto3’s Bedrock Runtime client to call Bedrock’s invoke_model with a model-specific request body.

The prompt format follows Llama 2 chat conventions, using “INST” with “human” and “assistant” sections. The request body includes generation controls such as max token length, temperature, and top_p. In the Lambda handler, the incoming API Gateway event is parsed to extract blog_topic from the request body, then generate_blog_using_bedrock is called to produce the blog text. The code reads the Bedrock response, pulls the generated content from the response JSON (the walkthrough references the generation field), and returns a success message.

After generation, the Lambda function writes the blog to S3. It creates a timestamped S3 object key (e.g., blog_output/<time>.txt) using the current time formatted as hours, minutes, and seconds. A dedicated helper function performs the S3 put_object call, storing the generated blog text in the configured bucket.

The integration layer is API Gateway (HTTP API). A POST route (blog generation) is created and integrated with the Lambda function, then deployed to a Dev stage. Testing begins with Postman; initial failures include a 404 “message not found” caused by a request path mismatch (including an extra space), and a Bedrock authorization error caused by the Lambda execution role lacking permission for bedrock:InvokeModel. Fixing the IAM role by attaching broader permissions resolves the invoke_model failure. After successful invocation, CloudWatch logs show the generated output, and the S3 bucket contains the timestamped .txt file with the blog content.