AWS Builder Center
Weekend Showcase Challenge: Daily Cloud Haiku

Weekend Showcase Challenge: Daily Cloud Haiku

An always-on agent that writes a fresh cloud/DevOps haiku every day — now extended with an on-demand mode that generates a haiku about AWS's "Service of the Day," built with EventBridge, Lambda, API Gateway, Bedrock, and S3.

Cloud Architect

Weekend Showcase Challenge: Daily Cloud Haiku

#application

Vision & What It Does

I'm a DevSecOps engineer, and most of my day is spent around logs, alerts, and infrastructure that quietly does its job in the background. Daily Cloud Haiku started from a simple question: what if that background noise could become something reflective instead?
The core agent runs completely on its own, once a day, with zero human input. It picks a random cloud/DevOps concept — a misconfigured S3 bucket, a Lambda cold start, an IAM policy, a database failover — checks the live weather in Lahore, Pakistan, and writes an original 5-7-5 haiku that's subtly shaped by both. The result gets published to a small public gallery that anyone can browse.
For this Showcase Challenge, I took the project further. Since the whole point of this weekend is showing what a summer of building added up to, I added a second, visitor-facing layer: a button on the gallery page that generates a fresh haiku on demand, themed around a rotating "AWS Service of the Day." Click it, and within a couple seconds you get an original haiku about that day's featured service — EC2, DynamoDB, EventBridge, whatever the rotation lands on. It's not stored anywhere; it's meant to be a small, ephemeral moment of creativity rather than another permanent gallery entry, and I think that restraint is part of the design.

How I Built It

The daily agent was already fully built and deployed from an earlier Summer Build Series weekend, so my job this time was refinement rather than starting over — exactly the kind of "add the finishing touch you ran out of time for" path this challenge calls out.
The biggest decision was keeping the two flows cleanly separated instead of bolting the new feature onto the existing scheduled Lambda. I built a second, independent Lambda function ( cloud-haiku-ondemand ) rather than modifying the original one. This meant zero risk of breaking the daily automated pipeline that was already working reliably, and it kept each function's responsibility narrow: one runs on a timer and writes to storage, the other runs on a request and returns a response.
The trickiest part was actually just process discipline, not code. Early on, a test invocation on the new Lambda returned in under 2 milliseconds — way too fast for a real model call. Checking CloudWatch Logs showed the function had never actually run my code; it was still the default "Hello World" stub because I'd pasted the code without clicking Deploy. It was a good reminder to always verify with logs, not just a "successful" test result, before assuming something works.
Once that was sorted, I matched the on-demand Lambda's Bedrock invocation pattern exactly to the proven, working pattern from the daily Lambda (same message body shape, same response parsing with a fallback for different provider response shapes), rather than guessing at a new format. That paid off immediately — the very first real test returned a clean haiku about Amazon EC2.
Wiring the frontend up was the last piece: a small vanilla JS  fetch()  call to an API Gateway endpoint, a loading state on the button while the request is in flight, and a result card that appears without a page reload.

AWS Services Used / Architecture Overview

Two flows share one static frontend:
Autonomous daily flow: Amazon EventBridge Scheduler (daily cron, 3 AM PKT) → AWS Lambda ( cloud-haiku-daily ) → fetches live Lahore weather → Amazon Bedrock (DeepSeek V3.2) generates the haiku → Amazon S3 stores the dated JSON entry and updates a rolling  latest.json  → the same S3 bucket serves the static gallery page.
On-demand AWS-theme flow (new): Visitor clicks a button on the gallery page → Amazon API Gateway (HTTP API,  POST /generate-haiku ) → AWS Lambda ( cloud-haiku-ondemand ) picks that day's AWS service from a fixed rotation → Amazon Bedrock (DeepSeek V3.2) generates a themed haiku → the JSON response is rendered directly in the browser.
Supporting both flows: AWS IAM roles scoped to least privilege (Bedrock invoke, S3 read/write for the daily function; Bedrock invoke only for the on-demand function), and Amazon CloudWatch Logs for verifying both scheduled and on-demand executions actually ran.

What I Learned (Across the Summer)

This summer's series pushed me from "I can deploy a Lambda function" to actually thinking in event-driven patterns — recognizing when something should be a scheduled trigger versus a request/response trigger, and not being afraid to add a second small function instead of overloading one. I also got a lot more comfortable debugging with CloudWatch Logs first, rather than trusting a green "success" checkmark at face value — the empty Lambda stub incident this weekend was a small but useful lesson in that. And building something that touches my actual daily work (cloud/DevOps concepts, styled by real conditions) made the whole series feel less like isolated exercises and more like a running project I actually want to keep extending.


Link to App / Repo

Shoutout

Tagging @Lewis Sawe, whose "The Museum That Grows" — an always-on agent that invents a new artifact for a fictional civilization every day — was the community spotlight from the last weekend's challenge and a great example of the same "always-on, no human in the loop" spirit this project leans on.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article