Bridge: A Serverless Communication Tool for Deaf Visitors at Service Counters
How I built a serverless communication bridge for deaf visitors at service counters using AWS Lambda, API Gateway, DynamoDB, and Polly - replacing the need for a scarce, pre-booked sign-language interpreter with an instant, on-demand tool
The problem
Walking into a bank, hospital, or government office is routine for most people. For a deaf visitor, it often isn't. Communicating with a teller, receptionist, or clerk usually requires a sign-language interpreter - and interpreters are scarce, need advance booking, and cost money. In the moment, at the counter, there's frequently no good option at all.
I wanted to solve this specific moment: not "accessibility" as an abstract goal, but the exact friction of one hearing person and one deaf visitor trying to communicate right now, with no interpreter in the room.
The idea
Bridge is a simple two-pane web app designed to sit on a tablet at a counter:
- The hearing person speaks normally. Their speech appears as live captions on screen for the visitor to read.
- The visitor types a reply. It's read aloud so the hearing person hears it without needing to read anything.
Whether or not the visitor can speak, they can type — so it works across the real range of deaf communication styles, not just one case. No booking, no waiting, no third person needed.
The architecture
Bridge runs entirely serverless on AWS:
Browser (live captions + typed reply) │ ▼ Amazon API Gateway │ ├──► Lambda: LogExchangeFunction ──► DynamoDB ├──► Lambda: GetSessionsFunction ──► DynamoDB └──► Lambda: SpeakReplyFunction ──► Amazon Polly- AWS Lambda runs three small functions - logging an exchange, retrieving session history, and converting text to speech - with no server to manage and no cost when idle.
- Amazon API Gateway exposes those functions as HTTPS endpoints the frontend calls directly.
- Amazon DynamoDB stores every exchange in real time - which doubles as a usable accessibility log for the institution.
- Amazon Polly converts the visitor's typed reply into natural speech, so the counter staff hears a real voice, not robotic text-to-speech.
- The whole stack is defined as code in a single SAM template and deployed with the AWS SAM CLI, an open-source tool that let me go from template to live infrastructure with two commands:
sam buildandsam deploy.
Live captions use the browser's built-in speech recognition rather than Amazon Transcribe streaming - Transcribe's browser-based streaming needs a Cognito identity pool for temporary credentials plus a duplex audio client, which is a larger, higher-risk build. I scoped that out deliberately in favor of a smaller, reliable core loop that actually works end to end.
What I learned
This was my first time deploying infrastructure end-to-end with SAM CLI, and my first time wiring a Lambda function to Amazon Polly for real-time text-to-speech inside a live application. The most useful lesson wasn't technical - it was scoping: cutting Cognito, multi-session dashboards, and Transcribe streaming early meant I had a working demo instead of five half-finished features.
Who this is actually for
The deaf visitor is the user, but the institution is the buyer. Under India's disability access regulations, banks and public offices are expected to provide reasonable accommodation for disabled customers. Bridge gives them a low-cost way to do that on demand - replacing the recurring cost of booking an interpreter with a one-time, reusable tool at every counter.
What's next
The clearest next step is regional language support - Hindi and Kannada captions and speech - since interpreter scarcity is worse for regional languages than for English in most Indian cities. Amazon Transcribe streaming is the other natural addition, once there's time to build the Cognito-backed audio pipeline properly rather than rushing it.
Built for the First Commit hackathon - Ship It track.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article