Building Generative AI Applications in 4 Hours
I shared how we built and deployed dtax, a voice medical triage prototype, during a four hour hackathon at Amazon.
Amazon · Singapore · 25 March 2026
Overview
We built dtax around one interaction: a person describes symptoms by voice, and the application prepares a structured handoff for a clinician. In my AWS User Group talk, I walked through how we moved from that idea to a deployed demonstration.
Four hours forced us to choose what to build and what to leave out. I focused on managed services, early testing, and deployment while there was still time to fix integration problems.
dtax was an educational prototype and was not clinically validated for patient care.
Workflow and Topics
Product workflow
We captured voice input and organized it into a triage summary, keeping the spoken account at the center of the handoff.
Delivery workflow
We scoped the core experience, built it, tested locally, and deployed. I used that cycle to explain our tradeoffs between time, cost, and output quality.
Architecture and Tools
I presented an AWS architecture covering Amplify, Bedrock, Cognito, DynamoDB, and Lambda. I also discussed Amplify versus Lightsail, a monolith versus microservices, and templates versus custom code.
My question for each choice was whether we could configure, test, and deploy it within the hackathon. Authentication, data modeling, and secrets still needed attention even in a small prototype.