The Saini Ledger

SEATTLE — Less tech-savvy users, often older relatives, get stuck on the web constantly: a confusing government portal, a login that keeps erroring, a form that will not submit, a scam page built to mislead. They rarely ask for help in the moment, and by the time family hears about it, the context is gone. The Frustration Detector was built for exactly that scenario — the grandchild's inbox reading: "They visited the login page four times, hit an error, and gave up on a form. View screenshot."

A Chrome extension watches for struggle signals locally. A content script scores DOM signals — clicks, scroll, form abandonment, idle time, cursor thrash, scam phrases — while a background script tracks repeated reloads, back-navigation, HTTP errors and known scam domains. Past a threshold, it captures a tab screenshot and ships the event out.

HOW IT WORKS

API Gateway hands events to a Lambda that validates and writes them to DynamoDB with a 30-day TTL, uploading screenshots to S3 as both a raw image and a self-contained HTML viewer.

Every 15 minutes, an EventBridge-scheduled Lambda re-scores events per URL and builds a plain-English summary. When the aggregate crosses the alert threshold, the backend mints a 7-day presigned URL to the screenshot viewer and publishes to SNS — emailing the designated helper which site the person was on and what happened. The whole stack is defined in AWS CDK and deploys in one command.

LESSONS FROM THE FLOOR

Two-stage scoring — instant locally, aggregated server-side — balanced responsiveness against false alarms; either threshold alone was too noisy. Privacy ran end to end: UUIDs instead of names, field masking, presigned links and short TTLs, essential when monitoring someone's screen for their own benefit.

Source available → github.com/SNBest1/AWS_hackathon ↗

CONTINUED: Banks try small models for personal offers, Page B3 →