// Case study · N°05

JoinRedeemer.

Chrome cashback extension across 30+ retailers. Cut GCP SQL costs by 42% through query and infra rework.

Role: Full-stack Engineer
Year: 2024 → 2025
Status: shipped
Stack: TypeScript · Node.js · GCP

Full-stack engineer on a US Chrome extension plus companion web app that activates cashback at checkout across major e-commerce sites and recommends the best credit card per purchase. I worked the extension (TypeScript + Redux Toolkit/RTK Query), backend integration, and the analytics pipelines. My main charge was integrating retailers at scale and killing the long tail of selector, cookie, redirect, and tracking bugs.

// Context

The extension runs in the background and "wakes up" only on shopping flows: it detects cart/checkout/confirmation pages via a per-retailer registry of selectors and URL regexes, redirects through an affiliate link, recommends a card, tracks the transaction, and reports to MixPanel and Everflow. Anonymous-first: a user gets an account on install with zero friction.

Scale

30+ retailers

Cost saved

42% GCP SQL (~$250/mo)

Testing

Jest → Vitest + Puppeteer

Anonymous-first

Retailer integration at scale .

Each retailer is a contract: cart/checkout/thank-you selectors plus URL regexes, SPA detection where needed, multi-region subdomain handling, then end-to-end validation of activation, tracking, and card recommendations in both visible and hidden extension modes. I integrated or maintained 30+: Amazon, Best Buy, Walmart, Home Depot, Nike, eBay, AliExpress, and a long tail, plus quarterly selector audits to keep the top retailers green.

Backend & infrastructure .

Anonymous → registered user merge with Everflow referrer/ID de-duplication.
Moved Everflow event tracking (install, sign-up, first purchase, uninstall) from frontend to backend with token-protected server-to-server postbacks and a Google-Cloud IP allowlist.
Built a static outbound IP (Cloud NAT + Serverless VPC Access Connector) so partner APIs could whitelist us.
Cut Google Cloud SQL fixed cost by 42% (~$250/month) by right-sizing after a usage analysis.
Replaced an external WhoisXML domain-classification call with a rolling 7-day top-25 domain table, cheaper and more controllable.

Testing .

Migrated the test runner Jest → Vitest and built a Puppeteer end-to-end pipeline that simulates real navigation to reach cart/checkout, deliberately bypassing the retailer bot-detection that blocks direct-URL access.