AWS Builder Center
Weekend Annoying Task Challenge: From CORS Nightmares to AI Success with Standup Sentinel

Weekend Annoying Task Challenge: From CORS Nightmares to AI Success with Standup Sentinel

Standup Sentinel is an AI powered developer productivity companion that transforms messy, unstructured daily notes into crisp, professional 3 point standup reports (What I Did, What I Will Do, Blockers). Built with React, AWS Amplify, AWS Lambda, and Amazon Bedrock (Nova 2 Lite).

Building the Future of Agentic AI | Full Stack Developer | Cloud Enthusiast

Vision & What the app does

Writing daily standup updates or weekly status reports is one of the most repetitive, low-value chores in software development. After a long day of debugging, reviewing PRs, and context switching, trying to remember and format exactly what you accomplished feels like pulling teeth.
I built Standup Sentinel to completely automate this annoying task. It is an AI powered developer companion that takes a messy, unorganized "brain dump" of notes and instantly converts them into a structured, professional 3 point standup update:
  1. What I did: Clean bullet points detailing completed work.
  1. What I will do: Planned tasks and upcoming priorities.
  1. Blockers: Current technical bottlenecks.
By pasting raw notes into the left panel, developers get a polished, team ready status update on the right panel in seconds saving mental energy and ensuring seamless communication.

How I Built It (The Journey & The Struggle)

I am relatively new to AWS, so I viewed this weekend challenge as the perfect excuse to dive into the deep end of serverless architecture. My goal was simple: build a React frontend and a Python AWS Lambda backend in 2-3 hours. I leveraged AWS Kiro's spec-driven development to rapidly scaffold the frontend components and draft the initial backend logic.
But building is rarely a straight line, and this is where the real learning began. Here is how I fought through the roadblocks:

1. The CORS 500 Error Nightmare

Once I deployed the React app on AWS Amplify and connected it to my Lambda Function URL, the app broke immediately. I was hit with a wall of generic  500 Internal Server Errors  masking themselves as CORS violations. I learned the hard way that when using a direct Lambda Function URL, if the Python script crashes internally, AWS automatically throws a 500 error, causing the browser to block the response due to missing headers. I had to learn how to dive into AWS CloudWatch logs to find the true cause.

2. The Bedrock Account Verification Block

CloudWatch revealed my first backend crash:  AccessDeniedException . Because my AWS account was new, it was placed under a mandatory verification hold. I learned that AWS places security holds on powerful services like Amazon Bedrock to prevent bot abuse.

3. The AWS Marketplace Payment Roadblock

After my account was verified, I tried to invoke Anthropic’s Claude 3 Haiku model. I was instantly hit with an  INVALID_PAYMENT_INSTRUMENT  error in my browser's Network tab. I realized that because Anthropic is a third-party model, AWS bills it through the AWS Marketplace, which strictly requires a traditional Visa/Mastercard credit card. My UPI AutoPay setup (which works for standard AWS services) was being rejected by the Marketplace.

4. The Final Pivot: Amazon Nova 2 Lite

Instead of scrambling to find a credit card, I pivoted my architecture. I realized that Amazon's native foundation models do not require AWS Marketplace billing.
I navigated to the Amazon Bedrock console, requested access to the newly released Amazon Nova 2 Lite ( us.amazon.nova-2-lite-v1:0 ), and updated my Python code.
I rewrote my  boto3  integration to use the newer, more robust Converse API (moving away from the fragile  invoke_model  payload structure).
Python
# My final working Converse API implementation
response = client.converse(
modelId="us.amazon.nova-2-lite-v1:0",
messages=[{"role": "user", "content": [{"text": f"Format these raw notes... \n\nNotes: {text}"}]}],
inferenceConfig={"maxTokens": 500}
)

I deployed the new code, refreshed my Amplify app, clicked "Format Report," and it worked flawlessly. The UPI payment setup was perfectly valid for Amazon's native model, and the Standup Sentinel came to life.

AWS Services Used / Architecture Overview

This application operates on a 100% serverless AWS infrastructure:
  • AWS Kiro: Used to generate the spec-driven architecture, scaffolding the React UI and drafting the Python backend.
  • AWS Amplify: Hosted my React frontend, providing seamless continuous deployment directly from my GitHub repository.
  • AWS Lambda (Function URLs): Acted as my serverless backend API. Exposing the function via a direct URL allowed me to bypass complex API Gateway configurations for this lightweight tool.
  • Amazon Bedrock (Nova 2 Lite): The brain of the operation. My Lambda function uses  boto3  and the  converse  API to pass user notes to the  us.amazon.nova-2-lite-v1:0  model for intelligent formatting.
  • AWS CloudWatch: Essential for debugging. I relied heavily on CloudWatch logs to trace execution errors, manage my Lambda memory footprints, and extend my function timeouts from the default 3 seconds to 30 seconds to accommodate the AI inference delay.

What I Learned

This challenge transformed me from a beginner into someone who genuinely understands how AWS serverless components communicate.
  1. Never Trust a Browser CORS Error: I learned that a frontend CORS error is often just a mask for a backend crash. Always check CloudWatch first.
  1. Bedrock Billing Nuances: I discovered the crucial difference between AWS Marketplace billing (third-party models like Claude) and standard AWS billing (native models like Nova).
  1. The Converse API is King: I learned that AWS's  converse  API is significantly cleaner and more reliable than passing raw JSON strings via  invoke_model .
I set out to automate an annoying task, but the real prize was walking away with a deep, battle-tested understanding of AWS architecture.

Link of App & Repo

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