Realtime streamer platform · Design · Backend
A freemium hub for Twitch streamers — a customizable identity page, and Twitch and Discord bots that push every event live to overlays. My flagship: the largest system I've designed, built and run solo, in production.
Streamers juggle a dozen separate services. Omiros pulls the important ones into one place: a bento page they make their own identity, community events like subgoals, and bots that react to what happens on stream in real time.
It's freemium, and it's the project where I forged the backend blueprint the rest of my work now reuses.
Twitch events arrive as EventSub webhooks. A Go core processes them. The pieces after that each use a different transport, chosen for its channel rather than for the buzzword.
Streamers can put arbitrary animation scripts into their overlays. So overlays are served from a separate, cookieless origin and rendered in a sandboxed iframe — a malicious overlay can't reach the session on api.omiros.me because the cookies simply aren't there. Most people would serve overlays from the same domain and get burned.
The same second-order thinking shows up elsewhere: banned words on public pages are blocked at save, so a troll can't get a streamer Twitch-banned through a page the streamer didn't even write.
When an overlay updates, the page reloads in full rather than patching the DOM. It's a blunt instrument — it flashes, and it kills smooth animation — and I chose it knowing that.
Overlays update rarely (a sub, a follow, a milestone), run unattended for hours, and a frozen overlay mid-stream is a public incident. A full reload means the overlay is consistent with the backend by construction, with almost no client state to drift. Robustness over elegance, with the limit named out loud.
Most of the four and a half months went into things you can't see on screen: multi-tenant overlay security, the Twitch EventSub lifecycle, a billing rebuild.
I also deleted a finished feature. Subathon shipped as a complete, tested module — then I removed it wholesale. Not because the code was bad, but because I couldn't answer how the user should experience it, and building backend behind an undecided UX is building waste. I'd rather leave an honest gap than a bad foundation something else comes to depend on.
I decide the architecture and enforce it; the code beneath is AI-written, and I read it back before it ships. Putting overlays on a cookieless origin inside a sandboxed iframe was a decision about untrusted code, not a snippet to generate — and understanding the output is what tells me it actually seals that boundary. Following the code closely is also how I catch a build heading the wrong way and steer it back before it drifts.
Three lines about your idea is enough. Reply within 24 hours.