Building.

The old app was a .NET WebView wrapper averaging 3.4 orders a day, sitting at ~1.8★. I owned the Flutter rebuild — Figma, onboarding UX, HTTP caching, performance, deploys. In 8 weeks: 17× daily orders (3.4 → 58), MAU 4.7K → 9.4K, ~11.6% session-to-purchase, <0.04% crash rate. What made it feel like an achievement wasn't the numbers — it was knowing real users were having a smooth experience where before they had frustration.

With the app stable, I turned to the web. Moving the legacy .NET frontend to Next.js (SSR/SSG/CSR) with TypeScript and a feature-based architecture — ~40% page-load reduction, 35% Core Web Vitals improvement, ~35% smaller JS payload. The web finally getting the same care the app did.

I joined Pazy as a frontend intern in December 2024 and was promoted to SDE by June 2025. Two kinds of work defined the time there: an OCR-powered reimbursement flow that deleted hours of mechanical finance work, and the quieter scaffolding around it — a Vue/Nuxt frontend, a design system, and a UTR reconciliation tool.

Before practicals, my class shared code through DontPad — plain text, no images, and anyone with the URL could overwrite anything. I built ShareBoard: a real-time board with a rich editor, images, password-protected read/edit modes. The whole class used it during finals. The name is the whole idea — a pad lets you write; a board lets you pin anything.

I move between LLMs a lot and lose context every time. FlowForge takes a long conversation and extracts the bones — goals, decisions, constraints, things I like, things I dislike — into a transferable file. The point isn't summarising; it's making context portable.

Most productivity tools block. I want friction. You tell it what counts as productive and unproductive for you. When you open something distracting, it doesn't stop you — it adds a five-second pause and asks why. The goal isn't control. The goal is awareness.

Archive tools show what a site was. I want to show what a site would have been. Feed it a 2024 product — it reimagines a 2008 version using design patterns, type, and constraints of that era. Part archive, part alternate history.