AWS Builder Center
Making Notices Understandable for All

Making Notices Understandable for All

A mobile-first AI app that turns complex Bengali and Telugu notices into plain language, extracts important actions and deadlines, and reads them aloud. Built for the WeMakeDevs Bharat Builds Tour, First Commit hackathon.

We thought the hard part was building it. It wasn't.

I went into the WeMakeDevs Bharat Builds Tour thinking the interesting part would be building an AI app on AWS.
Turns out, getting the AWS services to actually work was the interesting part.
The idea started from a pretty simple problem. Formal notices are everywhere, but a lot of them are written in language that is technically correct and practically difficult to understand. You can read Bangla or Telugu and still have no idea what a notice is actually asking you to do.
So we decided to build something around that.

The idea

We built Notice, made simple, a small mobile-first web app that takes a notice and turns it into something a normal person can actually understand.
The user can either paste the text or upload a photo. They choose their language and get:
  • A simple explanation of the notice
  • The important actions they need to take
  • Any deadlines mentioned
  • An audio version they can listen to
The goal wasn't to build another chatbot where you ask questions about a PDF.
We wanted the first screen to basically answer:
"What does this notice mean, and what do I need to do?"

Building it on AWS

The stack ended up being more complicated than the original idea.
The frontend is React with Vite and Tailwind, deployed through AWS Amplify. The backend runs on AWS Lambda, exposed through API Gateway.
For storage, we used S3 for images and generated audio, with DynamoDB for saved notices.
For the AI side, we used Amazon Bedrock. We experimented with both Claude and Nova depending on availability and account access.
OCR was a different story.
Textract worked for the languages it supports, but Bangla and Telugu created a problem. Instead of redesigning the entire project around that limitation, we packaged Tesseract into the Lambda environment and used it for those languages.
For audio, we ended up running Edge TTS from Lambda because Polly doesn't currently give us the Bangla and Telugu voices we needed.
So the final architecture wasn't exactly what we had on paper at the start.
It was whatever survived the weekend.

AWS had other plans

This was probably the most useful part of the entire hackathon.
At one point, Bedrock was just returning:
 Operation not allowed 
No matter what I changed.
Different regions. Different credentials. Different attempts at authentication. Same thing.
Then Textract started throwing:
 SubscriptionRequiredException 
And another teammate trying Claude got:
 INVALID_PAYMENT_INSTRUMENT 
At that point, spending another few hours trying to get one particular service working wasn't going to help us ship.
So we stopped treating the AWS errors as something we had to solve immediately and started treating them as engineering constraints.
We added region failover for Bedrock.
We added Groq as a fallback for text simplification.
We used Tesseract instead of depending completely on Textract.
We used Edge TTS instead of waiting for Polly support.
That changed how I think about cloud architecture.
A dependency isn't really a dependency if your entire demo stops working when it becomes unavailable.

The annoying details that actually mattered

The AI part looked impressive when we showed it.
The boring infrastructure was what made it actually work.
CORS had to be configured properly.
S3 permissions had to line up.
API Gateway had to accept the kind of payload we were sending.
Lambda environment variables had to exist in the deployed environment, not just on my laptop.
And the LLM had to return JSON that our frontend could actually parse.
That last one caused more problems than I expected.
If I asked the model for deadlines, sometimes I got:
{
"deadline": "2026-09-30"
}
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