
Weekend Showcase Challenge: Spark of the Day
An autonomous agent that wakes up once a day, invents a whimsical 120-word micro-story using Claude on Amazon Bedrock, and publishes it to a simple webpage — no human intervention required. My first AWS project, built for the Weekend Showcase Challenge.
Vision
Most days, I don't have time to write something creative — but I still like the idea of a little spark of imagination showing up on its own. That's what Spark of the Day does: it's a tiny autonomous agent that invents one short, whimsical micro-story every day, completely unattended, and publishes it to a simple webpage anyone can visit.
No prompts to type, no button to click. Every day it picks a random creative theme — a lighthouse keeper who talks to ships, a robot learning to garden, a shop that repairs broken memories — and asks Claude, running on Amazon Bedrock, to turn that theme into a short story under 120 words with a one-line takeaway at the end. The result gets saved and shown on a small, quiet, single-page site.
The problem it solves is small on purpose: a moment of unexpected creativity dropped into an otherwise ordinary day, generated by something that runs entirely on its own.

How I built it
This is genuinely my first AWS project, so I kept the scope small and the pieces simple:
-
agent.py— a Python script that calls Amazon Bedrock'sinvoke_modelAPI with a randomly chosen theme, gets back a short story from Claude, and writes it to a JSON file (plus a dated archive copy so nothing gets overwritten).
-
site/index.html— a minimal static page that fetches the latest JSON and displays the story with a bit of styling — no framework, just fetch and render.
The biggest challenge, unsurprisingly for a first project, was picking the right model ID. My first attempt used a Claude 3.5 Haiku model ID that had already reached end-of-life on Bedrock, which threw a
ResourceNotFoundException . From there I learned that newer Claude models on Bedrock are invoked through inference profiles rather than bare model IDs — so the working call ended up looking like:python
modelId = "us.anthropic.claude-haiku-4-5-20251001-v1:0"Once that was sorted, generating a story became a one-line
python agent.py away. I also managed to briefly break my own working setup by running a leftover sed command meant only for the failure case — a good reminder to always check why a fallback command exists before running it out of habit.AWS services and architecture
The current build is intentionally lean:
agent.py (Python, boto3) │ ▼Amazon Bedrock — Claude Haiku 4.5 (via inference profile) │ ▼site/latest.json + site/archive/YYYY-MM-DD.json │ ▼site/index.html (fetches and displays the latest spark)Right now the agent runs manually, with the plan to schedule it (via cron, or an AWS-native scheduler like EventBridge in a future iteration) so it truly runs itself daily without me remembering to run it.
What I learned across the summer
Coming into this challenge with zero prior AWS projects, the biggest lesson was just how approachable "agentic" builds can be once you break them into small pieces: one script, one API call, one JSON file, one HTML page. I also learned the practical difference between a model ID and an inference profile ID on Bedrock, why model lifecycle matters when picking an ID to hardcode, and the value of testing a script's happy path before wiring up any fallback or error-handling commands around it.
Mostly, I learned that "ship something small and real" beats "plan something big and never finish it" — which felt like the right note to end the summer series on.
Repo
Tagging a builder who inspired me
Big shoutout to Lewis Sawewhose Weekend Creative Agent Challenge: The Museum That Grows — an always-on agent that invents one new artifact a day for an imaginary civilization's museum — was the direct inspiration for the "runs on its own, once a day" shape of this project. Give them a follow.
#application
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article