Weekend Annoying Task Challenge: StandupDrafter
Turning messy end-of-day notes into a clean, ready-to-post status update — automatically.
Vision & What the App Does
Every week I lose time to the same small chore: turning a scrappy list of things I did — half-finished notes, a meeting here, a bug fix there — into something coherent enough to post as a status update. It's not hard work, but it's exactly the kind of low-value repetitive task that eats a chunk of a Monday morning.
StandupDrafter is a small web app that fixes that. You paste in rough, messy notes — "fixed login bug, met with sarah, started api docs, blocked on staging access" — and it hands back a clean, structured update organized into Done, In Progress, and Blockers, ready to paste straight into Slack or email. It also keeps a running history of your past updates, so you can glance back at what you reported last week without digging through old messages.
The goal wasn't to build something complicated. It was to take one annoying five-minute task and make it a five-second task.


How I Built It
I used Kiro to go from idea to a fully specced, fully implemented app over the weekend. Kiro generated a requirements document and a design document before writing any code, which turned out to be genuinely useful — it caught scope questions (how should history be scoped per user? what happens on empty input?) before I'd written a line of infrastructure.
From there, Kiro broke the build into 54 discrete tasks covering the CDK infrastructure, the Lambda backend, the frontend, and a full test suite (infrastructure tests, frontend unit tests, and responsive design tests across mobile/tablet/desktop breakpoints).
The build itself went smoothly. Deployment was where the real engineering happened. A few things I had to debug hands-on:
- A 405 error on every request — traced to the frontend calling the S3 website URL instead of the actual API Gateway endpoint, because the API base URL wasn't being injected into the HTML at build time. Fixed by explicitly setting it in
index.htmlbefore the app script loads.
- A Bedrock
ValidationException— my first instinct was that the AI model call was misconfigured. After working through IAM permissions and cross-region inference profile ARNs, it turned out the real issue was that Bedrock model invocation was restricted at the account level (an anti-abuse measure that also applies to AWS credit/education program accounts). I confirmed this by testing multiple models directly in the Bedrock Playground — all of them returned the identical error, ruling out a code or permissions problem.
- The pivot — rather than lose the weekend waiting on an access request, I swapped the Bedrock call for a lightweight rule-based categorizer in the Lambda function: it parses the notes and buckets each item into Done, In Progress, or Blockers using keyword heuristics, then formats the output exactly the way the AI version would have. The app works identically from a user's perspective — the intelligence just lives in Lambda logic instead of a model call.
AWS Services Used / Architecture Overview
- Amazon S3 — hosts the static frontend (HTML/CSS/JS) as a public website
- Amazon API Gateway — REST API exposing
/generateand/historyendpoints
- AWS Lambda (Python 3.12) — processes notes into a formatted update and reads/writes history
- Amazon DynamoDB — stores each user's update history, keyed by user ID and timestamp
- AWS IAM — least-privilege roles scoping the Lambda function to only the DynamoDB actions it needs
- AWS CDK — defines and deploys the entire stack as infrastructure-as-code, so the whole app stands up with a single
cdk deploy
How it's triggered end-to-end: a user opens the S3-hosted site, types notes, and clicks Generate. The frontend calls the API Gateway
/generate endpoint, which invokes the Lambda function. Lambda parses and categorizes the notes, writes the result to DynamoDB alongside a timestamp, and returns the formatted update to the browser. The /history endpoint follows the same path in reverse — Lambda queries DynamoDB for that user's past entries and returns them to populate the history list.What I Learned
- CDK deployment debugging is a different skill from writing CDK code.
cdk synthpassing tells you your template is valid — it tells you nothing about whether your frontend is calling the right URL or whether your IAM policy covers every ARN pattern a service actually needs at runtime.
- Bedrock cross-region inference profiles have real IAM implications — granting
bedrock:InvokeModelon a single-region foundation model ARN isn't enough if the inference profile can route to other regions.
- Account-level service restrictions are worth ruling out early. When every model across every provider fails identically in the Bedrock Playground using my own credentials — not just the Lambda's IAM role — that was the signal it wasn't a permissions bug at all.
- A working fallback beats a blocked ideal. The rule-based version isn't as flexible as an LLM-backed one, but it's real, deployed, and solves the actual problem — which matters more than the tech stack looking impressive on paper.
Link to App
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article