
Weekend Creative Challenge: Deploy Log Bard
An AI-powered app that turns CI/CD deploy logs into epic sagas — triumphant if the build succeeded, tragic if it failed. Built with Amazon Bedrock Nova Lite, AWS Lambda, API Gateway, and S3, with real debugging stories on cross-region inference profiles and CORS.
Every DevOps engineer knows the feeling: you stare at a wall of CI/CD log
text, trying to figure out — in one glance — whether the deployment was
a triumph or a disaster. Deploy Log Bard takes that same wall of
text and turns it into something worth reading: a short epic saga,
written in the style of old heroic poetry, that tells the story of your
deployment. Succeed, and you get a triumph. Fail, and you get a tragedy
— but a beautifully written one.
This is
a serverless, AI-powered app running entirely on Amazon Bedrock, AWS
Lambda, Amazon API Gateway, and Amazon S3.
Vision & what the app does
As a Cloud Engineer working daily with AWS, Kubernetes,
Terraform, and CI/CD pipelines, I wanted a Creative Challenge
entry that didn't just generate a generic poem from a mood or a prompt
— I wanted it to be genuinely rooted in my own world. Deploy Log Bard
takes a raw deployment log — a
terraform apply output, a Kubernetes
rollout, a GitHub Actions run, a Jenkins pipeline, anything — and:- Parses it for real signals: success or failure, pipeline stages, duration, and the specific line that sealed the outcome.
- Sends those facts (not the raw log) to Amazon Bedrock's Nova Lite model with a prompt asking for a short epic saga in that tone.
- Returns a titled saga with a Triumph or Tragedy badge, styled like an illuminated manuscript. The result: paste in a real outage log, and instead of dread, you get a tiny dramatic retelling of your own incident — genuinely funny when you're not the one who has to fix it.
How I built it — and what actually went wrong
I wanted this article to be honest about the build, because the real
value for other builders is in the debugging, not just the demo.
Architecture decision. I kept this deliberately simple: one Lambda
function does log parsing and the Bedrock call, fronted by an API
Gateway HTTP API, with the frontend as a single static HTML file on S3.
No database, no auth, no over-engineering — just enough AWS services to
tell a clean architecture story.
Issue #1 —
pip install aws-sam-cli failed with
externally-managed-environment . My WSL/Ubuntu environment runs
Python 3.12 with PEP 668 protections that block system-wide pip
installs. Fix: installed pipx instead ( sudo apt install pipx ), which
gives command-line tools like SAM CLI their own isolated environment
without touching system Python packages — the correct tool for
installing a standalone CLI application rather than a project
dependency.
Issue #2 —
sam build "succeeded" but silently built nothing. My
first sam build reported Build Succeeded , but buried in the output
was Skipping copy operation since source ... lambda does not exist .
The lambda/ and frontend/ subfolders referenced in template.yaml
didn't actually exist yet — I'd saved my files flat in the project root
instead. A reminder that SAM's "success" message only means the build
step ran without erroring, not that it packaged real code. Fixed by
creating the correct folder structure and verifying file sizes with
ls -la before rebuilding.Issue #3 — Bedrock
ValidationException : on-demand throughput not
supported. After a clean deploy, my first real API test returned:"Invocation of model ID amazon.nova-lite-v1:0 with on-demand throughput isn't supported. Retry your request with the ID or ARN of an inference profile that contains this model."
This is a regional nuance worth knowing: in
ap-southeast-1
(Singapore) and other APAC regions, Amazon Nova Lite requires a
cross-region inference profile ID, not the bare model ID. I found the
correct one with:aws bedrock list-inference-profiles --region ap-southeast-1 \
--query "inferenceProfileSummaries[?contains(inferenceProfileId,'nova-lite')]"which returned
apac.amazon.nova-lite-v1:0 . Swapping the Lambda's
BEDROCK_MODEL_ID environment variable to this inference profile ID —
via sam deploy --parameter-overrides BedrockModelId=apac.amazon.nova-lite-v1:0
— fixed it immediately.
Issue #4 — Frontend worked via curl but failed in the browser with
"Failed to fetch." Direct
curl calls to the API succeeded, but the
browser-based frontend on S3 couldn't reach it. Classic CORS
problem: the browser sends a preflight OPTIONS request before the
real POST , and my HTTP API had no route or headers configured to
handle it. Fixed by adding an explicit CorsConfiguration block under
Globals: HttpApi: in template.yaml , allowing POST / OPTIONS and
the Content-Type header, then rebuilding and redeploying.
After changing thr CORS, the final result!!!


]
AWS services used / architecture overview
- AWS Lambda (Python 3.12) — parses the deploy log for outcome and key facts, then orchestrates the Bedrock call
- Amazon Bedrock (Nova Lite, via APAC cross-region inference profile) — generates the epic saga from structured facts
- Amazon API Gateway (HTTP API) — public POST endpoint with CORS enabled for browser access
- Amazon S3 — static website hosting for the frontend
Browser (S3 static site)
│ POST log text
▼
API Gateway (HTTP API, CORS-enabled)
│
▼
AWS Lambda (Python)
├─ parses log → outcome, stages, duration, key line
└─ calls ──────────────────────────► Amazon Bedrock (Nova Lite)
│
returns saga title + text
◄───────────────────────────────────────┘
│
▼
JSON response → rendered in browserDeployed entirely with AWS SAM (Serverless Application Model) —
sam build + sam deploy — for reproducible infrastructure-as-code.What I learned
- Regional model access nuances matter. Not every Bedrock model is
invocable via its plain model ID in every region — APAC regions in
particular route Nova models through cross-region inference profiles.
Always check
aws bedrock list-inference-profilesbefore assuming aValidationExceptionmeans your account lacks access.
- "Build Succeeded" isn't proof of a correct build. Always check the SAM build log for skipped copy operations, not just the final status line.
- CORS is still CORS, even for HTTP APIs. SAM's
HttpApievent source doesn't auto-configure CORS — it has to be declared explicitly underGlobals.
- Grounding the model in extracted facts, not raw logs, produced noticeably better output — the saga stayed accurate to real pipeline stages and error lines instead of drifting into generic fantasy text.
Link to app or repo
Built solo over a weekend for the AWS Weekend Creative Challenge using
Amazon Bedrock, AWS Lambda, Amazon API Gateway, Amazon S3, and AWS SAM.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article