Univerze

Live streaming, end to end. I brought LiveKit into the codebase and built the rooms, then the economy around them: virtual coins bought through Stripe, gifting, creator payouts, live battles, and later captions that turn speech into text with a short AI summary of what the stream is about.

The feed. I wrote the first ranking algorithm (interest scoring, then topic classification and student recommendations) and the XP and milestone system that rewards activity.

Money and growth tools. Stripe-funded jobs, milestones and hire offers for the freelance marketplace; a shop with seller payouts; and most of the internal sales CRM, from CSV lead import and Gmail reply sync to tracked outreach, a public demo-booking page and in-browser calling over Twilio.

AI products on the OpenAI API: a website builder with a live Sandpack preview, an app builder previewed in Expo Snack, a workflow-automation canvas, and an auto-reply stand-in for people who are offline.

I set up the GitHub Actions deploy to EC2 in the first week and added Sentry later. When a deploy wiped the production build and the site returned 504s, I traced it through the PM2 logs, stopped the crash loop and restored it from CI. I also fixed builds that ran out of memory on the runner, and schema drift between production and staging that crashed the feed.

By October the codebase had about 5,700 files. I first authored roughly 28% of them, about 450 of the 1,220 API route handlers and 200 of the 570 pages, alongside one other lead engineer and several other contributors. In September I also delivered an employer and school mentorship expansion: a 41-table schema, more than 100 routes, scheduled jobs and an end-to-end test.