SDE-2 at Loop Health — building the enrolment and benefits platform for 1M+ members · Pune, India · open to remote
I went from intern to SDE-2 in 24 months by owning problems end to end: the React frontend a member clicks through, the Node.js services behind the API Gateway, the pipeline that ships both, and the dashboards that page me when something breaks. Health insurance is unforgiving about data — a wrong dependent or a double checkout is someone's coverage — so I care a lot about idempotency, observability and boring, reliable releases.
📈 Most of my commits are in a private organisation. The graph below is real activity — ~1,500 contributions this year across 8 repositories — you just can't click through to the code. Ask me about any of it.
- Building the corporate benefits enrolment portal at Loop Health — React 18 + 3 Node.js services on Cloud Run, serving clients of 800 to 8,000 employees
- Shipping psyMing, an offline-first PWA for habits and reminders, on evenings and weekends
- Contributing to speaktype, a 100% offline voice-dictation app (#142)
- Exploring how far agent rules and MCP can take AI-assisted engineering on a real production codebase
Core
Backend
Cloud
Platform & Delivery
Reliability & Security
AI Engineering
| | |
| Per-PR preview environments | Every pull request gets a usable product link in under 60 seconds, auto-updated on each commit. Replaced a multi-day wait on shared staging for product, design and QA. |
|---|---|
| Deploys: 10 min → under 1 min | Parallelised GitHub Actions, partial-clone checkouts, pinned toolchain, Slack release reporting. Migrated dev and prod pipelines to keyless Workload Identity Federation. |
| Observability from scratch | 21 custom Datadog RUM vitals and the team's production health dashboard: p95 latency and error rate by endpoint, Core Web Vitals by route, errors-per-session by release. Frontend MTTR 2 days → under 1 hour. |
| Web performance | Login load time −67% (4.5 s → 1.5 s), initial JavaScript −93% (3.3 MB → 230 KB), production build −61%. Lighthouse 60 → 90. Landed the tree-shaking work in the shared component library that every client app now imports. |
| Idempotent checkout | Designed the state-transition guards after a race between async benefit writes and checkout confirmation could wipe member benefits. Checkout is replay-safe. |
| 20,000-row table on a laptop | Operations console member view from 2 GB → 40 MB of heap with row virtualization. Usable on low-end operator machines. |
| MCP for the codebase | An internal Model Context Protocol integration that streams live browser UX context into AI coding agents, plus the repository agent rules that keep AI-assisted changes on team conventions. |
| Component library | #1 contributor to the company's shared React library. 1,000+ pull requests reviewed across 8 repositories. |
- psyMing — an offline-first PWA for habits, reminders and milestones. React, Redux Toolkit, MUI and localforage on the front; Express + Mongoose on Vercel Functions behind it, with web push driven by a cron scheduler. Built because every habit app I tried needed a network connection to tell me to drink water.
- speaktype — 100% offline, open-source voice dictation. Fixed the control pill and menu-bar panel so they render over full-screen apps (#142).
- Stacked PRs, small diffs. One reviewable idea per PR; 1,000+ reviews taught me what unreviewable looks like.
- Instrument before you optimise. Every perf number above came after a vital or a dashboard existed to measure it.
- AI in the loop, conventions in the repo. Agent rules and MCP context live next to the code so AI-assisted changes look like the team wrote them.
- Async by default. Written specs, decision records, and PR descriptions a stranger could act on.
Portfolio: gauravlonari.github.io · Writing: freeqry.blogspot.com · Email: gauravlonariofficial@gmail.com