The day job
- 90%smaller Docker images for Next.js on AWS ECS, with half the memory
- 19 to 3minutes for the median test run
- 30%faster Cutshort, after I led the move from AngularJS to Next.js
- 20%more interaction from real-time chat and Kanban boards
Recent wins
Faster builds, lighter containers
Rebuilt the frontend's CI and Docker pipeline: caching between CI jobs, a base image with the dependencies already in it, shared packages cached in their own layers, and a multi-stage production build. Images went from about 1.4 GB to 0.12 GB, so containers start faster on ECS, and build artifacts from about 1.4 GB to 0.06 GB.
Ten times fewer calls on every page
Took a user lookup out of the Next.js middleware of the busiest portal, where it ran on every request. Calls to the user service fell from about 320,000 a day to about 30,000, and the portal's Apdex score rose from 0.68 to 0.72.
A quarter off a database bill, with one setting
Worked out that the audit trail's cost was mostly stored history that is rarely read, and moved its DynamoDB tables to the Standard-IA class. That class charges more for every read and write, so a quiet table alone doesn't justify it; what decided it was that the saving on the stored table and its indexes outweighed the dearer requests. Its database spend fell 25% the next month, an estimated 33% at the same usage, with no change to the application. It was recognised in the quarter's achievements.
Half the wait on a busy API
Cut the average latency of the API behind the portals' task cards in half. I also traced a slow-response problem to how traffic was spread across the servers, with the data to show it.
A new regulator onboarded without custom code
Shipped client-based verification: verification flows that are set up by configuration rather than written for each client. It let a national nursing and midwifery council come on board without any custom work.
Monitoring that costs what it's worth
Found that nearly all of our browser monitoring bill came from about 40 million resource and request events a month that nobody used. I turned those off and kept what we rely on: JavaScript errors, Web Vitals and page timings. Failed requests were already caught by our API monitoring, so nothing we used was lost.
The Dataflow Group
Senior software engineer, Bengaluru, India, now
- Implemented a multi-layer Docker pipeline for Next.js deployments on ECS, reducing image size by 90% and memory usage by 50%.
- Optimized the test runner from a 19 minute median to a 3 minute median.
- Built a browser-based PDF editing feature with drag-and-drop, rotate, and reordering workflows.
- Created an automated document verification layer that pre-validated submissions against requirements and reduced rejection rates plus human review load.
- Built a lead creation platform with custom CRM integrations, Calendly scheduling, and complex payment flows.
- Mentored junior engineers and guided an intern through end-to-end delivery of a PWA with geotagging, browser-based image manipulation, and Google Maps integration.
Cutshort
Frontend engineer, Remote, before
- Improved site performance by 30% by leading a migration from AngularJS to a modular Next.js App Router architecture.
- Enhanced user interaction by 20% by implementing WebSockets for real-time chat and building Kanban board experiences.
- Refactored large legacy files into reusable components and implemented Optimistic UI and Infinite Queries for a more seamless experience.
Want the long version? Ask me.
Write to the page